Deprescribing in Older People Approaching End of Life

A Randomized Controlled Trial Using STOPPFrail Criteria

Denis Curtin, MB; Emma Jennings, MB; Ruth Daunt, MB; Sara Curtin, MSc; Mary Randles, MB; Paul Gallagher, PhD; Denis O'Mahony, MD


J Am Geriatr Soc. 2020;68(4):762-769. 

In This Article



This study was a parallel-group unblinded randomized pragmatic clinical trial conducted in two acute hospitals in Cork City, Ireland. Participants were randomized to receive usual pharmaceutical care (ie, hospital physician and pharmacist care) or usual pharmaceutical care supplemented by individualized STOPPFrail-guided deprescribing. The local Clinical Research Ethics Committee approved the trial protocol. The trial was registered with (NCT03501108).


Eligible participants were hospitalized older adults (aged ≥75 y), admitted from the community with acute unselected medical or surgical illness, who following treatment were unable to return home to independent living and consequently required long-term nursing home care. Eligible participants were prescribed five or more long-term medications and were severely frail as defined by (1) a Clinical Frailty Scale[17] score of 7 or higher, and (2) the treating physician indicating that he or she "would not be surprised if the patient died in the next 12 months."[18,19] Patients were excluded if they, or, in the case of cognitively impaired individuals, a proxy was unwilling or unable to provide informed consent.

Comprehensive multidisciplinary long-term nursing home care applications are reviewed every 2 weeks at a local placement panel meeting presided over by a consultant geriatrician. These applications, which include details about diagnoses, medications, and functional and cognitive status, were used to screen for potentially eligible participants. Patients with a Mini-Mental Status Examination (MMSE) score of 24 or higher were considered competent to provide written informed consent.[20] For patients with a diagnosis of dementia or those with an MMSE score lower than 24, a nominated proxy was required to cosign the consent form. The full trial protocol can be found in Supplemental Protocol S1.

Data Collection

A trained research physician (D.C.) conducted patient and/or caregiver interviews and medical record reviews to collect the following baseline data before randomization: (1) current and past diagnoses; (2) long-term regular medications and pro re nata (PRN; as needed) medications (PRN medications recorded if used three or more times in the previous week); (3) functional status (modified Barthel Index[21]); (4) comorbidity status (Charlson Comorbidity Index[22]); and (5) quality of life (QUALIDEM[23] and ICECAP-O[24]). In addition, symptoms such as pain, sleep disturbance, and gastrointestinal symptoms were explored in an unstructured manner by the research physician. After baseline data collection was completed, the research physician used the STOPPFrail criteria to identify deprescribing targets. Medications targeted for deprescribing were recorded in the case report form.

Quality of life (QoL) was measured using two validated assessment tools. Anticipating that a large proportion of participants would have advanced dementia and therefore could have difficulty completing self-reported questionnaires, the QUALIDEM instrument was selected.[23] It is completed by nursing staff or healthcare assistants and assesses QoL across multiple domains for people at all stages of dementia.[23] In addition, participants, where possible, or a proxy were requested to complete the ICECAP-O questionnaire, a broad measure of QoL (ie, beyond health) that was developed for use in the economic evaluation of health and social care interventions in older adults.[24] Both the QUALIDEM and ICECAP-O were previously used to measure QoL in institutional care settings.[25,26]


Participants were randomized to study arms in a 1:1 ratio using block randomization. Block sizes of four and six were generated using the website ( by an administrator external to the study. Randomization was not stratified by hospital site. The allocation sequence was concealed in sequentially numbered opaque envelopes until the research physician had enrolled participants, completed baseline data collection, and identified deprescribing targets using the STOPPFrail criteria.


For participants randomized to the intervention arm, a medication withdrawal plan was devised by the research physician. The recommended medication withdrawal plan was communicated directly to one of the participant's attending physicians and also documented in the patient's medical record. Medications associated with an increased risk of an adverse withdrawal reaction were recommended to be withdrawn slowly according to a standardized trial withdrawal protocol (Supplemental Protocol S1). The attending physician judged whether or not to accept the drug withdrawal plan and implement the recommended changes. Because of the nature of the intervention, the research physician, attending physicians, and participating patients could not be blinded to the intervention or control group assignment after randomization. The intervention was applied at a single time point during the patients' hospital admission at the time of trial enrollment. Attending physicians and nursing staff were encouraged to report any potential adverse consequences of deprescribing (adverse drug withdrawal events or disease relapse) to the research team.

Outcome Measures

The primary outcome was the mean change in the number of long-term prescribed medicines consumed by participants at 3 months after randomization. Short-term medicines (eg, antibiotics, topical antifungal agents, topical corticosteroids) were not included. For combination products, each ingredient was included as one drug as long as that ingredient was available as a single medicine in the contemporaneous British National Formulary, 74th ed.[27]

Secondary outcomes were measured at 3 months and included the following:

  1. Unscheduled medical reviews and emergency transfers after discharge from the acute hospital

  2. Falls and nonvertebral fractures after discharge from the acute hospital

  3. Changes in prescriptions of neuroleptic antipsychotic medications

  4. Changes in 28-day cost of participants' prescription medications

  5. Changes in participants' QoL (measured by the QUALIDEM instrument and the ICECAP-O questionnaire)

  6. Mortality

Outcome data were collected by three research physicians (E.J., R.D., and M.R.) who were blinded to the group allocation of participants. We contacted directors of nursing in the relevant nursing homes by telephone and requested them to complete a case report form populated with the relevant data fields. We requested that a nurse or care assistant, familiar with the participant, complete the QUALIDEM instrument. Where possible, the ICECAP-O was to be completed by the same person who completed the questionnaire at baseline. In some instances, the research physicians contacted the relevant person by telephone to complete the ICECAP-O. We calculated the 28-day cost of participants' prescription drugs using a 2018 Irish pharmaceutical wholesaler price list, produced by Clanwilliam Health. For each specific medication dose and formulation, the lowest cost option was chosen.

Sample Size Calculation and Statistical Analysis

We calculated the statistical power of the trial to detect a difference of 2 in the mean number of medications between the intervention and control groups (α = .5; 1-β = .8; population variance = 14 [taken from a recent prevalence study[28]) at 3 months. Allowing for an estimated attrition rate (deaths and dropouts) of 30%, we estimated that a sample size of 160 participants (80 in each group) would be required.

In the analysis of the primary outcome, we included only participants who completed follow-up. Because medications regimens frequently change in the final stages of terminal illness, we excluded deceased participants due to difficulties in determining final valid, verifiable medication lists. Emergency department presentations, hospital admissions, and mortality were determined on all randomized participants. We used standard descriptive statistics with study groups compared using χ 2 or Fisher exact tests for categorical variables, the independent samples t test for normally distributed continuous variables, and the Wilcoxon rank-sum test for nonparametric variables. We performed statistical analyses using SPSS v.25.