Introduction

An important principle when caring for older people with multi-morbidity is to carefully align the medication regimen to the condition and goals of care of the individual patient [1]. This is particularly important for patients approaching end of life where symptom management usually takes priority over stringent chronic disease control. Polypharmacy is common in this cohort and many of these patients are prescribed medications that are probably futile [2]. Yet, physicians commonly forego the opportunity to deprescribe because of fear of negative consequences (i.e., symptom relapse, clinical deterioration) [3, 4]. This is despite evidence indicating that deprescribing can be achieved without compromising patient safety or wellbeing [5,6,7].

The complexity associated with frailty, multi-morbidity, and polypharmacy necessitates a systematic approach to deprescribing. Scott and colleagues have recently proposed a 5-step deprescribing protocol (CEASE—confirm current medications, estimate risk of drug-related harm, assess each medication for discontinuation, sort/prioritize medications for discontinuation, eliminate medications according to agreed deprescribing plan). The third step—assessing each medication for discontinuation—requires the user to answer a series of questions about each medication in the patient’s regimen (Fig. 1) [8]. While comprehensive and patient-centered, the outcome of this step will depend on the knowledge, attitudes, and experience of the user. Implicit approaches, such as CEASE, are usually time-consuming, thereby greatly limiting their integration into routine clinical practice [9]. More recently, the STOPPFrail criteria (Table 1), a list of 27 indicators to assist physicians with deprescribing decisions in frail older individuals with poor 1-year survival prognosis, have been validated [11]. Of the 27 indicators, 26 are explicit (i.e., clearly defined statements highlighting the potentially inappropriate use of particular drug/drug classes in a particular clinical situation) and one is implicit (i.e., A2: stop any drug without a clear clinical indication). STOPPFrail criteria, which are organized according to physiological system, are concise, have substantial inter-rater reliability [12], and are designed to be used by physicians of all disciplines who provide care for frailer older people on a routine basis.

Fig. 1
figure 1

Step 3 of the CEASE protocol: Scott’s deprescribing algorithm

Full size image
Table 1 The STOPPFrail criteria [10]
Full size table

The primary aim of the present study is to compare the utility of the structured predominantly explicit STOPPFrail criteria with a gold standard comparator in frail older people with poor 1-year survival prognosis. Of the available published deprescribing guides, the CEASE protocol has the strongest evidence of efficacy and physician acceptability [10], and therefore, its use by a physician with expertise in clinical pharmacotherapy is an appropriate gold standard for deprescribing. If STOPPFrail reproduces the results of this gold standard, then its brevity and easy usability may make it a more appropriate method of deprescribing in routine clinical practice for this particular population of older people. The secondary aim was to determine which inappropriate or unnecessary medications are not identified by STOPPFrail. This information could inform future iterations of the STOPPFrail criteria.

Methodology

Clinical cases

To ensure that the comparison between the two deprescribing methods was valid, it was important to minimize external sources of variability [13]. For this reason, structured clinical cases were prepared to ensure timely and equal access to information relevant to the deprescribing decision (Supplementary appendix 1). These clinical cases were based on anonymized patients included in a recent observational study that examined the prevalence of potentially inappropriate medications in the discharge prescriptions of older people hospitalized in the year prior to their death [2]. Each structured clinical case included a list of diagnoses, regular medications, functional and cognitive status, and routine blood tests results prior to hospital discharge. All clinical cases were based on patients aged ≥ 65 years, prescribed ≥ 5 regular medications with moderate to severe frailty (Clinical Frailty Score ≥ 6 [14]). For each of the clinical cases, it was assumed as follows:

  1. i.

    The patient was medically stable

  2. ii.

    The patient had a poor 1-year survival prognosis

  3. iii.

    The list of diagnoses was complete and correct

  4. iv.

    Laxatives (unless potentially part of a prescribing cascade) and paracetamol were appropriate

  5. v.

    There were no difficulties with medication administration (e.g., dysphagia, poor inhaler technique) unless explicitly stated

  6. vi.

    The patient’s nutritional status was satisfactory unless otherwise stated

  7. vii.

    Behavioral and psychological symptoms of dementia were present only if explicitly stated

Application of deprescribing methods

Four physicians, all trained in geriatric medicine, reviewed the clinical cases and identified medications that were potentially eligible for deprescribing. Two physicians (DC and DOD) rigidly applied STOPPFrail criteria while the other physicians (KJ and TD), who were not familiar with STOPPFrail criteria, identified drugs to be deprescribed using step 3 of the CEASE protocol (hereafter referred to as Scott’s deprescribing algorithm; Fig. 1). The physicians were instructed to document the primary reason for each deprescribing decision. Drugs that were not eligible for deprescribing were classified as “important.” The physicians initially worked independently and then resolved any discrepancies in pairs to produce a final consensus list for each deprescribing method.

Sample size calculation and statistical analysis

A sample size of 100 was chosen to detect with 80% probability a Cohen’s kappa coefficient of 0.70 under the alternative hypothesis when Cohen’s kappa under the null hypothesis was 0.6. This sample size would also allow for more than 500 medications to be evaluated. Cohen’s kappa coefficient was interpreted as poor if ≤ 0.2, fair if 0.21–0.40, moderate if 0.51–0.6, substantial if 0.61–0.8, and almost perfect if 0.81–1.00 [15]. Statistical analysis was performed using SPSS® version 21.

Results

Clinical cases

The mean number of medications per clinical case was 10.2 (standard deviation 3.3). The total number of medications to be evaluated (when paracetamol was excluded) was 994. Most medications were taken orally (88.7%), while the remainder were administered by inhaled (5.1%), transdermal (3%), topical (2%), or subcutaneous/intramuscular (1.3%) routes.

Agreement between methods

The physicians using Scott’s deprescribing algorithm identified 524 medications (52.7% of the total) as potentially eligible for deprescribing; the physicians using STOPPFrail criteria identified 412 medications for deprescribing (41.4%; see Supplementary appendix 2). Cohen’s kappa coefficient was 0.60 (95% confidence interval 0.55–0.65; p < 0.001) indicating moderate agreement between the methods. With Scott’s deprescribing algorithm representing the gold standard, the sensitivity of STOPPFrail (i.e., the proportion of inappropriate medications correctly identified by STOPPFrail) was 70.2%. The specificity (i.e., the proportion of important medications that were correctly continued by the physicians using STOPPFrail) was 90.6%. The positive predictive value of STOPPFrail (i.e., the proportion of medications deemed inappropriate by the physicians using STOPPFrail that were actually inappropriate) was 89.3% while the negative predictive value (i.e., the proportion of medications deemed important by the physicians using STOPPFrail that were actually important) was 73.2%.

The primary reasons for the deprescribing decisions are summarized in Supplementary appendix 3. “No valid indication” was the primary reason for 50% of the deprescribing decisions made by the physicians using Scott’s deprescribing algorithm and in 42.7% of the decisions made by the physicians using STOPPFrail. Lipid-lowering agents, proton pump inhibitors, calcium, and anti-resorptive drugs for osteoporosis accounted for 33% of the medications deprescribed using STOPPFrail.

Discrepancies between methods

The physicians using STOPPFrail did not identify 156 medications (29.7%) that were potentially eligible for deprescribing (Table 2). Antihypertensive agents, vitamin D supplements, and laxatives (prescribed as part of a prescribing cascade) accounted for 54.4% of the potentially inappropriate medications that were not identified by the physicians using STOPPFrail. The physicians using STOPPFrail deprescribed calcium supplements and continued vitamin D preparations in all cases while the physicians guided by Scott’s algorithm were more selective and generally continued these medications when a history of osteoporosis, fractures, or recurrent falls was included in the patients’ medical history.

Table 2 Discrepancies between the deprescribing methods. STOPPFrail-guided deprescribing evaluated against “gold standard” deprescribing
Full size table

Discussion

This study is important because it shows that approximately half of all the medications prescribed to older people approaching end of life may be unnecessary or inappropriate. Many people with advanced frailty and polypharmacy will not have the benefit of a comprehensive specialist medication review. In this study, application of STOPPFrail—a novel, concise explicit deprescribing tool designed for all physicians who commonly provide care for older adults approaching end of life—demonstrated moderate agreement with gold standard specialist geriatrician-led deprescribing.

A major barrier to deprescribing is the difficulty associated with balancing risk and benefit of a specific medication for a particular patient. STOPPFrail addresses this difficulty by explicitly highlighting circumstances where commonly used medications can be safely discontinued. There is good evidence that people are much more likely to follow through on tasks that they see value in when those tasks are made easier for them [16,17,18]. It is therefore likely that providing explicit criteria will make the task of deprescribing more accessible to non-specialist physicians who care for older adults approaching end of life.

The physicians using the STOPPFrail criteria identified 70.2% of medications that were potentially eligible for deprescribing according to gold standard assessment. When medications for deprescribing were identified by the physicians using STOPPFrail, these medications were actually inappropriate in 89.3% of cases. While the use of STOPPFrail does not “catch all” potentially inappropriate medications, it is very reassuring that the great majority of the deprescribing decisions appear to align with gold standard care.

For both methods, the most common reason for deprescribing was “no valid indication.” This emphasizes the importance, during a medication review, of ensuring that each drug is linked to a diagnosis or active symptom. While STOPPFrail explicit criteria largely address step 2 (harm outweighs benefit) and step 4 (preventive drugs—benefit unlikely to be realized) of Scott’s deprescribing algorithm, future iterations may need to go further to address aspects of step 3 (symptom or disease control drugs). For example, STOPPFrail does not prompt the physician to review symptoms such as pain which may be over-treated with potentially problematic medications. Furthermore, symptoms such as poor appetite, nausea, altered bowel habit, sedation, and gait disturbance, which may represent the adverse effects of drugs, are not targeted. Finally, antihypertensive therapies and vitamin D supplements were the most common inappropriate or unnecessary medications that were not identified by the physicians using STOPPFrail. These drugs are commonly prescribed yet evidence of clear benefit, as well as specific guidance for use in people with advanced frailty, is lacking [19,20,21,22]. In the absence of high-quality clinical trial evidence, explicit criteria based on expert consensus opinion may enable physicians to make clinically sound decisions about the use of these medications in this particular expanding patient population.

All structured clinical cases in this study were derived from data collected from a cohort of hospitalized patients who died within 1 year of their hospital admission. A CFS score ≥ 6 was used to select frail patients from this cohort which would ensure that the deprescribing task was credible and that a short-term risk of death was not unforeseeable. It is important to emphasize that, in everyday clinical practice, we do not recommend using a CFS score ≥ 6 to select patients for STOPPFrail-guided deprescribing. STOPPFrail is intended for older people approaching end of life for whom the goal of care is to enhance quality of life and minimize the risk of drug-related complications. In the absence of sensitive and reliable prediction models [23], identifying older people who are approaching end of life will depend largely on physician experience and judgment [11].

Our study has some potential limitations. Firstly, it was a theoretical exercise using structured clinical cases. While derived from real patient data, the structured clinical cases do not reflect the complexities and nuances of real clinical care. However, we contend that standardization was necessary because external sources of variability (e.g., inequality of information) could have invalidated the primary aim of the study which was to compare the two methods of deprescribing [13]. Secondly, two physicians trained in geriatric medicine, arriving at deprescribing decisions through consensus, using Scott’s deprescribing algorithm, represented “gold standard” deprescribing in this study. It is important to emphasize that “gold standard” does not necessarily mean “perfect” but rather “best available” [24]. We believe the method used in this study is likely to be very close to the “best available” deprescribing for this population of patients in most hospitals.

In summary, the results of this study indicate that the STOPPFrail criteria can assist physicians in making appropriate deprescribing decisions and that, reassuringly, these decisions align closely with gold standard deprescribing. In everyday clinical practice, where frail older people approaching end of life are commonly encountered by attending physicians with variable expertise, STOPPFrail-guided deprescribing may be a reasonable alternative to specialist medication review. Future iterations of STOPPFrail should include guidance on antihypertensive therapy discontinuation as well as prompts to the physician to explore particular symptoms which may represent adverse drug events.