Discontinuation of medications at the end of life: A population study in Belgium, based on linked administrative databases

Abstract

Aims

The aim of this study was to examine the use of potentially inappropriate medication (PIM) in relation to time before death, to explore whether PIMs are discontinued at the end of life, and the factors associated with this discontinuation.

Methods

We conducted a retrospective register‐based mortality cohort study of all deceased in 2012 in Belgium, aged at least 75 years at time of death (n = 74 368), using linked administrative databases. We used STOPPFrail to identify PIMs received during the period from 12 to 6 months before death (P1) and the last 4 months (P2) of life.

Results

Median age was 86 (IQR 81–90) at time of death, 57% were female, 38% were living in a nursing home, and 16% were admitted to hospital between 2 years and 4 months before death. Overall, PIM use was high, and increased towards death for all PIMs. At least one PIM was discontinued during P2 for one in five (20%) of the population, and 49% had no discontinuation. Being hospitalized in the period before the last 4 months of life, living in a nursing home, female gender and a higher number of medications used during P1 were associated with discontinuation of PIMs (respective aOR [95% CI]: 2.89 [2.73–3.06], 1.29 [1.23–1.36], 1.26 [1.20–1.32], 1.17 [1.16–1.17]).

Conclusion

Initial PIM use was high and increased towards death. Discontinuation was observed in only one in five PIM users. More guidance for discontinuation of PIMs is needed: practical, evidence‐based deprescribing guidelines and implementation plans, training for prescribers and a better consensus on what inappropriate medication is.

1.

What is already known about this subject

• Many medications can be considered as inappropriate at the end of life.

• Discontinuation of potentially inappropriate medications (PIMs) in frail older adults with a limited life expectancy may improve medication use at the end of life, reduce adverse drug reactions and negative health outcomes, and support and improve quality of life.

What this study adds

• PIM use was high and increased towards death.

• For one in five of the population, at least one PIM was discontinued in the last 4 months of life.

• Hospitalization, living in a nursing home, female gender and the number of medications were associated with discontinuation of PIMs

2. INTRODUCTION

Managing medication use in people suffering from advanced stages of a life‐limiting disease is very challenging. In accordance with the definition of palliative care, care goals at the end of life should shift from quantity to quality of life.1 This should be reflected in medication prescription and use near the end of life. Adequate medication use in this situation means treating symptoms which are currently undertreated, as well as preventing possible harm caused by potentially inappropriate medications (PIMs).

PIM use has been studied in older adults with a normal life expectancy. Implicit (e.g. MAI2) and explicit criteria (e.g. Beers,3 STOPP/START4) have been developed and validated, aiming to identify PIM use in this population, and to assist physicians with deprescribing these PIMs. However, some medications considered to be inappropriate in the general older population may be used appropriately—eg for symptom relief—in a palliative care setting. Thus, these criteria require adaptation in order to be applicable in palliative care.5

In advanced stages of a life‐limiting disease, medication for symptom relief is often combined with medication to treat life‐limiting diseases, co‐morbidities and medication for long‐term prevention.6 However, many of these medications can be considered as potentially inappropriate at the end of life.7, 8 Moreover, some of these PIMs are often involved in drug–drug interactions with medications for symptom relief.9, 10, 11

Recently, explicit criteria to identify PIM use in frail older adults with limited life expectancy (STOPPFrail) were developed and validated.12 Due to age‐related changes in pharmacokinetics and pharmacodynamics, and a high prevalence of polypharmacy and inappropriate prescribing, frail older adults are more susceptible to adverse drug reactions (ADRs) and related negative health outcomes such as hospitalizations.12, 13, 14 Discontinuation of PIMs in frail older adults with a limited life expectancy may improve medication use at the end of life, reduce ADRs and negative health outcomes, and support and improve quality of life. It is crucial to get an insight into the current prescribing and use of PIMs in this population to determine the need for guidance in this area, eg for development of clinical practice deprescribing guidelines and interventions to reduce PIM use.

For this study, discontinuation is considered as an umbrella term for tapering or stopping PIMs in the specific context of limited life expectancy, eg by deprescribing those PIMs. Discontinuation of anti‐hypertensives, benzodiazepines, neuroleptics and statins has been associated with physical and cognitive benefits, and no significant harm in patients with a life‐limiting disease.15, 16 However, research has demonstrated that the diagnosis of a life‐limiting disease has little effect on the use and continuation of these PIMs.6, 17, 18, 19, 20, 21

This retrospective register‐based mortality cohort study aims (1) to get an insight into PIM use according to STOPPFrail in relation to time before death in a large population of deceased in 2012 in Belgium, (2) to explore to what extent PIMs are discontinued at the end of life, and (3) to examine the factors associated with discontinuation of PIMs.

3. METHODS

3.1. Study design and population

We conducted a retrospective register‐based mortality cohort study of people aged 75 years or older at time of death, who died in Belgium in 2012 and were registered by one of the seven healthcare insurers, which is mandatory for all legal residents.

3.2. Data source

Death certificate data, census data and fiscal data were obtained from Statistics Belgium, and were deterministically linked at the individual level to the InterMutualistic Agency's (IMA) national registry of healthcare claims data of the seven healthcare insurers in Belgium, and to the Belgian Cancer Registry. The resulting database covers approximately 99% of the full population who died in 2012. All databases were linked in a secure and ethically responsible manner to guarantee anonymity of the deceased. More information on the different databases, the linking procedure and the data protection approvals was published elsewhere.22

3.3. Assessment of outcomes

All medication data in this study were dispensing data from all hospital and community pharmacies, for all medications that were prescribed by any physician and reimbursed by the seven healthcare insurers in Belgium. Healthcare insurance is legally mandatory in Belgium, so data are complete for all legal residents, and registered in the IMA database. However, data on dispensing over‐the‐counter medications are not included in the database. Medications were classified based on the World Health Organization's ATC classification,23 which divides drugs into different groups according to the organ or system on which they act and their chemical, pharmacological and therapeutic properties. ATC codes at all levels—from one (anatomical main group) to five (chemical substance)—of every medication were traceable in the database.

We selected PIMs available on the Belgian market and listed on the STOPPFrail list of explicit criteria for PIM use in frail older adults with limited life expectancy, for which no specific patient‐level clinical information was needed to determine inappropriateness.12 Based on experts' opinions (R.V.S. and T.C.) and the available evidence, we categorized these PIMs into three groups: medications for long‐term prevention, medications for which chronic use is inappropriate, and (outdated) medications for which a safer alternative exists. Box 1 provides a more detailed description of the selected PIMs and our categorization.

BOX 1 Potentially inappropriate medications according to STOPPFrail selected for this study.

STOPPFrail criteria 12

STOPPFrail criteria 12

PIMs for long‐term prevention Lipid‐modifying agents Calcium supplements Osteoporosis drugsa SERMS for osteoporosis PIMs for which chronic use is inappropriate Memantine Sex hormones Neuroleptic antipsychotics Proton pump inhibitors (PPIs) H2‐receptor antagonists Gastrointestinal antispasmodics Long‐term oral steroids (Outdated) PIMs for which a safer alternative exists Theophylline Long‐term oral NSAIDs 5‐alpha reductase inhibitors Alpha‐blockers

Medications not selected because clinical information is needed to determine their appropriateness Anti‐platelets Leukotriene antagonists Muscarinic antagonists Diabetic oral agents ACE inhibitors for diabetes Angiotensin receptor blockers Prophylactic antibiotics Medications not available because they are not reimbursed in Belgium Multi‐vitamin combination supplements Nutritional supplements

^a Osteoporosis drugs include: bisphosphonates, bisphosphonate combinations, bone morphogenic proteins, other drugs affecting bone structure and mineralization (eg strontium).

We defined discontinuation as no dispensing during the last 4 months of life (P2) of selected PIMs that were dispensed at least twice during the period 12–6 months before death (P1). We defined initiation as no dispensing of PIMs during P1, and at least one dispensing during P2. Although our aim was to examine to what extent PIMs were discontinued at the end of life, we added data on new initiation of PIMs to counterbalance our results on discontinuation and situate these results in a proper context. P1 is identified in the database as day 365 to day 180 before death, and P2 as day 120 before death to day of death (Figure 1). All data on the prevalence of PIMs were based on dispensed prescription data of reimbursed medications from hospital and community pharmacies registered in the IMA database.

Figure 1.

Timeline representing the different time periods using the example of an individual dying on 30 June 2012, we portray the different time periods used in this article on a retrospective timeline

3.4. Measurement of individual characteristics

Study participants' age at time of death and gender were derived from the IMA database. Other socio‐demographic characteristics at time of death, such as household type, highest attained educational level, net taxable income and urbanization, were obtained through record linkage at the individual level with the socio‐demographic dataset and socioeconomic survey from Statistics Belgium. The household category “collective household” includes mainly nursing homes. To identify those with a cancer diagnosis, the Belgian Cancer Registry was linked to the other databases.

Characteristics on healthcare use during the period of 2 years to 4 months before death, such as hospitalization, visits by family physician, specialist palliative care and legal palliative care status, were derived from the IMA database. In this article, we refer to this period as the period before the last 4 months of life. Individuals received “specialist palliative care” when they were admitted to a palliative care unit in hospital or consulted a specialist multidisciplinary palliative home care team. Individuals acquired “legal palliative care status” after being diagnosed by a physician as suffering from advanced irreversible disease, with poor prognosis, and expected death in a relatively short frame.24

3.5. Data handling

For every selected PIM, the corresponding ATC code was selected from the IMA database. In Belgium, each prescription of medications is valid for 3 months. Therefore, the prevalence of a specific PIM during P1 is counted as the percentage of people for whom this specific PIM was dispensed at least twice during this period. If this specific PIM was dispensed only once during P1 or not at all, it was not counted. For the prevalence during P2, the specific PIM had to be dispensed only once to be counted.25 Thus, we can distinguish four groups in our population for every selected PIM: (1) a group for whom a specific PIM is dispensed at least twice during P1 and not dispensed during P2 (= discontinuation), (2) a group for whom a specific PIM is dispensed at least twice during P1 and at least once during P2 (= continuation), (3) a group for whom a specific PIM is not dispensed during P1 and dispensed once during P2 (= initiation), and (4) a group for whom a specific PIM is not dispensed during P1 nor during P2 (those who never used this PIM). A dichotomous variable was constructed to distinguish people for whom at least one of the selected PIMs was discontinued from those without discontinuation. This variable was used as the outcome for the logistic regression analyses. All other individuals—not belonging to any of these two subgroups—were excluded from further analyses.

To count the number of chronic medications during P1, every fifth level ATC code that was dispensed at least twice was counted. For the number of medications dispensed during P2, every dispensed fifth level ATC code was counted. PIMs were included in the number of medications and were not counted separately.

3.6. Statistical analyses

All statistical analyses were performed using Statistical Analysis Software (SAS^®) 9.4 and SAS^® Enterprise Guide 7.1 (SAS^® Institute Inc., North Carolina, USA).We used descriptive methods to describe the characteristics of the study population, use, discontinuation and initiation of PIMs. In a sensitivity analysis, people who died from sudden and possibly unexpected causes were excluded, but this rendered no meaningful differences in population characteristics and outcomes, so the general 75+ population was retained for analysis. A logistic regression model was used to examine the factors which were independently associated with discontinuation of PIMs. The variables considered for the multivariable logistic regression were those considered to be clinically important, ie those for which a 5 percentage point difference was found between the different categories of those variables in the univariate analyses. For continuous variables, a mean difference of at least three was considered clinically important. We adjusted the model for the remaining covariates.

3.7. Ethics

Data were anonymized. In accordance with Belgian law, approvals for access to the various databases and the database integrating all databases were obtained from two separate national sectoral committees for privacy protection: the Sectoral Committee of Social Security and Health, Section Health and the Statistical Supervisory Committee. Both are subcommittees of the Belgian Commission for the Protection of Privacy. In addition, the ethics committee of Ghent University Hospital provided approval (B670201422382).

3.8. Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY.26

4. RESULTS

4.1. Characteristics of the study population

Overall, 74 368 deceased individuals—median age 86 (IQR: 81–90) at time of death, 57% female—were included in this study. As shown in Table 1, 38% were living in a nursing home, and 23% were diagnosed with cancer. During the period before the last 4 months of life, 16% were admitted to hospital, with a median stay of 5 days (IQR: 3–7).

Table 1.

Characteristics of the study population and subgroups

	All deceased ≥75 years n = 74 368	At least one PIM discontinued n = 14 395	No discontinuation of PIMs n = 36 696	Others (not included in the two subgroups) n = 23 277
Age in years at time of death median (IQR)	86.0 (81–90)	85.0 (81–89)	85.0 (81–89)	87.0 (82–91)
Gendera (%):
Male	43.3	40.7	46.8	39.3
Female	56.7	59.3	53.2	60.7
Household typea (%):
Single person	25.1	23.6	28.2	21.1
Couple with no children living at home	27.1	27.3	31.1	20.4
Couple with children living at home	3.9	3.8	4.4	3.3
Single parent family	4.0	3.6	4.5	3.3
Nursing homeb	37.9	39.9	29.7	50.2
Unknown	1.9	1.8	2.2	1.8
Highest attained educational level (%):
No education	8.5	9.1	8.5	8.2
Primary education	37.5	38.2	37.6	36.9
Lower secondary	21.0	21.5	21.2	20.4
Education upper secondary	11.3	11.3	11.6	10.8
Education higher education	7.2	6.9	7.6	6.8
Unknown	14.5	13.1	13.5	16.9
Net taxable income (%):
< €10.000	23.4	23.9	23.7	22.8
€10.000–€15.000	27.1	27.2	26.6	27.9
€15.001–€20.000	26.8	27.2	27.1	25.9
> €20.000	22.7	21.7	22.6	23.4
Urbanization category (%):
Low	12.3	12.5	12.2	12.2
Middle	25.7	25.8	26.4	24.5
High	27.6	27.7	27.8	27.1
Very high	31.3	31.9	31.3	30.9
Unknown	3.1	2.1	2.2	5.3
Cancer diagnosis (%)	23.3	27.2	27.8	13.7
Hospitalization between 720 and 121 days before deatha (%)	16.0	25.5	8.8	21.5
No. of days: median (IQR)	5.0 (3–7)	6.0 (4–9)	4.0 (3–7)	4.0 (3–7)
Visits by family physician between 720 and 121 days before death:
median (IQR)	16.0 (7–26)	19.0 (10–30)	17.0 (8–26)	14.0 (4–23)
Specialist palliative care:
Onset specialist palliative care: median (IQR) days before death	22.0 (6–85)	36.0 (8–128)	18.0 (5–58)	28.0 (5–189)
Onset of specialist palliative care (%)
Never	85.4	81.3	83.2	91.5
Very early onset	0.5	1.0	0.3	8.5 (720–121)
Later onset (120–0)	14.1	17.7	16.5	8.1
Legal palliative care status
Onset legal palliative care status:
Median (IQR) days before death	38.0 (11–122)	56.0 (16–158)	31.0 (10–86)	42.0 (8–187)
Onset of legal palliative care status (%)
Never	89.4	85.5	88.2	93.8
Very early onset (720–121)	2.1	3.5	1.7	1.7
Later onset (120–0)	8.5	11.0	10.1	4.5

^aVariables with at least 5 percentage point difference or mean difference of at least 3 between the group with at least one PIM discontinued and no discontinuation of PIMs.

At least one PIM discontinued: at least one of the selected PIMs was discontinued between the period of 12–6 months before death (P1) and the last 4 months of life (P2).

Specialist palliative care was defined as being admitted to a palliative care unit in hospital or receiving palliative care at home from a specialist multidisciplinary palliative home care team. Legal palliative care status in Belgium is acquired after being diagnosed by a physician as suffering from advanced irreversible disease, with poor prognosis, and expected death in a relatively short term.23

^bCollective household including mostly nursing homes, long‐term care institutions for disabled persons, jail.

4.2. Potentially inappropriate medication (PIM) use during P1 and P2

In the total population (n = 74 368), the mean number of dispensed chronic medications was 6 (SD 4.86) during the period of 12–6 months before death (P1) (see Table 2). Most prominent PIMs for long‐term prevention during P1 were lipid‐modifying agents (21.5%). In the group of PIMs for which chronic use is inappropriate, proton pump inhibitors (PPIs) (28%) and neuroleptic antipsychotics (14%) were most common, and in the group of outdated PIMs, long‐term oral non‐steroidal anti‐inflammatory drugs (NSAIDs) were most prominent (7%).

Table 2.

Prevalence, discontinuation and initiation of dispensed potentially inappropriate medications (PIMs) (%) according to STOPPFrail (n = 74 368)

Medications	P1 n = 74 368	P2 n = 74 368	Discontinuation % (n/N)	Initiation % (n/N)
No. of medications mean (SD)	6.38 (4.86)	18.86 (11.79)	NA	NA
PIMs for long‐term prevention
Lipid‐modifying agents	21.5	25.1	21.1 (3383/16 016)	4.6 (2421/52 582)
Calcium supplements	4.8	11.3	40.8 (1443/3539)	7.6 (5145/67 858)
Osteoporosis drugs	6.3	8.3	28.5 (1337/4687)	2.9 (1969/67 620)
SERMS for osteoporosis	0.2	0.1	34.4 (42/122)	0.01 (6/74 208)
PIMs for which chronic use is inappropriate
Memantine	1.1	1.0	31.6 (269/851)	0.2 (156/73 411)
Sex hormones	0.9	1.0	29.2 (192/656)	0.4 (284/73 632)
Neuroleptic antipsychotics	14.4	31.4	16.7 (1790/10742)	20.6 (12 311/59 646)
Proton pump inhibitors (PPIs)	28.2	51.8	8.9 (1869/20993)	31.8 (15 081/47 416)
H2‐receptor antagonists	6.3	13.9	28.6 (1346/4709)	9.2 (6194/67 122)
Gastrointestinal antispasmodics	4.1	26.6	41.2 (1256/3049)	24.3 (16 185/66 632)
Long‐term oral steroids	9.9	29.3	22.2 (1642/7390)	21.9 (13 458/61 471)
(Outdated) PIMs for which a safer alternative exists
Theophylline	2.1	2.8	18.0 (276/1536)	0.8 (602/72 298)
Long‐term oral NSAIDs	6.8	15.8	47.0 (2386/5079)	11.6 (7202/62 102)
5‐alpha reductase inhibitors	1.6	4.5	36.0 (437/1214)	3.1 (2226/72 251)
Alpha blockers	0.2	0.8	30.8 (36/117)	0.6 (460/74 103)

P1 = 12–6 months before death (denominator: total population ≥75); P2 = the last 4 months of life (denominator: total population ≥75); D = Discontinuation of PIMs (P1 = 1 and P2 = 0) = within the group taking PIMs at 12–6 months before death, prevalence of discontinuation of these PIMs the last 4 months of life (denominator: total population ≥75 for whom P1 = 1), I = Initiation of PIMs (P1I = 0 and P2 = 1) = within the group taking no PIMs at 12–6 months before death, prevalence of initiation of these PIMs the last 4 months of life (denominator: total population ≥75 for whom P1I = 0).

The number of dispensed medications increased to 19 (SD 11.79) during the last 4 months of life (P2). The prevalence of all PIMs increased, more specifically to 25% for lipid‐modifying agents, 52% for PPIs, 31% for neuroleptic antipsychotics, and 16% for NSAIDs.

4.3. Discontinuation of PIMs

Between P1 and P2, at least one selected PIM was discontinued for one in five (20%) (n = 14 395) of the population. No discontinuation of PIMs was observed for 49% (n = 36 696). People for whom at least one PIM was discontinued had a median age of 85 years (IQR: 81–86), 59% were female, 40% were living in a nursing home, and 26% were hospitalized in the period before the last 4 months of life, with a median stay of 6 days (IQR: 4–9). For those without discontinuation of PIMs, median age was 85 (IQR: 82–91), 53% were female, 30% were living in a nursing home, and 9% were hospitalized in the period before the last 4 months of life. The mean number of chronic medications used during P1 was 9.6 (SD 5.71) for those with at least one selected PIM discontinued compared to 5.7 (SD 4.23) for those with no discontinuation of PIMs (data not shown).

4.4. Discontinuation of PIMs vs initiation of new PIMs

Among the users of the selected PIMs during P1, the percentage of discontinuation varied between 8.9% for PPIs and 47% for long‐term NSAIDs. The prevalence of newly initiated PIMs during P2 varied between <1% for selective estrogen receptor modulators (SERMS), memantine, sex hormones, alpha blockers and theophylline and 32% for PPIs. The prevalence of discontinuation exceeded the prevalence of initiation for theophylline, lipid‐modifying agents, osteoporosis drugs, H2‐receptor antagonists, sex hormones, alpha blockers, memantine, SERMS, 5‐alpha reductase inhibitors, calcium, gastrointestinal antispasmodics and long‐term NSAIDs. For PPIs and neuroleptic antipsychotics, the prevalence of initiation exceeded the prevalence of discontinuation. For long‐term oral steroids, the prevalence of discontinuation and initiation were equal (see Figure 2).

Figure 2.

Prevalence (%) of discontinuation vs initiation of the selected PIMs during the last 4 months of lifeFor every selected PIM, the percentage of discontinuation is represented on the x‐axis and the percentage of new initiation on the y‐axis. Medications in the red zone of this figure are of particular concern at the end of life due to the high prevalence of new initiation and low prevalence of discontinuation during the last 4 months of life

4.5. Factors associated with discontinuation of PIMs

For the variables “being hospitalized within the period before the last 4 months of life”, “living in a nursing home”, and “female gender”, a 5‐percentage point difference was found between the categories at least one PIM discontinued and no discontinuation of PIMs in the univariate analyses. For the mean number of chronic medications during P1, a mean difference greater than three between the two subgroups was found. These variables were considered to be clinically important and their association with discontinuation of at least one PIM was examined in the multivariate logistic analysis. We controlled for age, educational level, net taxable income, urbanization, cancer diagnosis, visits by family physician, specialist palliative care and legal palliative care status. The odds of discontinuation of PIMs increased significantly in association with the variables “being hospitalized within the period before the last 4 months of life”, “living in a nursing home”, “female gender”, and a higher number of chronic medications used during P1 (respective aOR [95% CI]: 2.89 [2.73–3.06], 1.29 [1.23–1.36], 1.26 [1.20–1.32], 1.17 [1.16–1.17]).

5. DISCUSSION

5.1. Key findings

Overall, PIM use according to STOPPFrail was high during the last year of life and increased towards death. Apparently, physicians continue to prescribe medications that are potentially inappropriate until the very end of life. Probably, prognostic uncertainty plays an important role here, as well as a lack of consensus on which medications are inappropriate at the end of life. The prevalence of PIMs during both time periods differs for each PIM; figures can be found in Table 2. In the group of PIMs for long‐term prevention, the prevalence of lipid‐modifying agents was high. For these medications, discontinuation exceeded new initiation in the last 4 months of life. In the group of medications for which chronic use is inappropriate, the prevalence of PPIs and neuroleptic antipsychotics was high. Furthermore, in both therapeutic groups new initiation exceeded discontinuation in the last 4 months of life. The prevalence, discontinuation and new initiation of outdated medications was more limited.

For one fifth of the population, at least one PIM was discontinued in the last 4 months of life, while no discontinuation of PIMs was observed for nearly half of the population (see Table 3). People who were admitted to hospital during the period before the last four months of life, nursing home residents, and women had more chance of discontinuation of PIMs. People for whom at least one PIM was discontinued during the last 4 months of life used more chronic medications during the period 12–6 months before death.

Table 3.

Factors a associated with discontinuation of potentially inappropriate medications (PIMs)

^aVariables with a ≥5 percentage point difference or mean difference ≥3 between the category with at least one PIM discontinued and no discontinuation of PIMs in univariate analyses were considered to be clinically important (circled in red). At least one PIM discontinued: at least one of the selected PIMs was discontinued between the period of 12–6 months before death (P1) and the last 4 months of life (P2). Model was adjusted for age, education level, net taxable income, urbanisation, cancer diagnosis, visits by family physician, specialist palliative care, legal palliative care status.

5.2. Strengths and limitations

We used a population‐level linked database with detailed demographic, socioeconomic and healthcare use information on all decedents in 2012 in Belgium. This allows us to follow back the dispensing of reimbursed prescribed medications up to 1 year before death. Although only services covered by insurers are included, in Belgium, where healthcare insurance is mandatory, data are relatively complete for healthcare services in hospital, nursing home and home settings.22 Consequently, we were able to examine PIM use and discontinuation of PIMs in the full population of those deceased at age 75 years and older. The number of 74 368 decedents at age 75 years and older is in accordance with the Belgian population statistics.27 To the best of our knowledge, this is the first study to include the balance between discontinuation of PIMs and initiation of new PIMs, which adds some refinement to the current picture of discontinuation of medications at the end of life.

The use of hospital and community pharmacy dispensing data to determine PIM use has certain limitations. First, as the data are based on reimbursed dispensed medications, we have to rely on the assumption that patients who received these medications also take them. In accordance with studies on compliance using administrative databases,25, 28, 29 we defined discontinuation as “at least two” dispensing during P1, and no dispensing during P2, to counterbalance this limitation. Second, administrative data are generally coarse grained: in Belgium, prescribed medications are dispensed for 3 months, and neither data on prescribed daily dose nor number of days of supply were available for this study. Consequently, we cannot examine concomitant use of medications, which complicates assessing their appropriateness. Third, some of the selected PIMs are difficult to observe in the IMA database because of their availability over‐the‐counter (e.g. NSAIDs, calcium). Prevalence of discontinuation might be overestimated for these PIMs if patients only get their first prescription filled at the pharmacy and afterwards buy these medications over‐the‐counter. However, these PIMs are more expensive when bought over‐the‐counter. Hence, the influence of this limitation on our results is likely to be minimal. Fourth, due to inaccessibility of clinical patients’ data, only PIMs for which no clinical information is needed to determine whether their use is inappropriate or not—14 out of the 26 on the STOPPFrail list—were included in the analyses. Thus, the high prevalence of PIMs overall in this study is likely an underestimation. However, research has demonstrated that criteria for which clinical information is not required can be reliably used to identify PIMs with a structured screening tool such as STOPP,30 whose applicability is comparable to STOPPFrail. The absence of clinical patient information complicates interpretation of our findings and does not allow for estimation of the treatment risk–benefit ratio of a specific medication for a specific patient, for example. Moreover, given we had no access to clinical patient‐level information, we were not able to adjust our multivariate model for comorbidities. Thus, residual confounding is possible. Finally, the relatively low prevalence of discontinuation must be interpreted with caution, as healthy‐user/sick‐stopper bias and prognostic uncertainty are common in a population aged 75 and older.31

With these strengths and limitations in mind, we can identify dispensing of PIMs during well‐defined time periods, and draw cautious, coarse‐grained conclusions regarding their discontinuation. Research has demonstrated that the main determinant of PIM use is the number of prescribed medications,14, 32, 33 which is probably confounded by the number and type of co‐morbidities.33 Adding clinical patient data and data on concomitant use of medications and treatment duration would create extensive opportunities for further research.

5.3. Interpretation in the context of literature

Overall, when death approached, prescribers continued treatment as before: PIM use was high at both time points, and increased towards death. Few changes in prescribing patterns were observed in relation to time before death: discontinuation and new—probably symptom‐driven and therefore not necessarily negative—initiation of PIMs was very limited. In a palliative care context, initiation of some of these PIMs may be indicated and likely to benefit the patient at the end of life, eg initiation of haloperidol to treat delirium when death is imminent, or chemotherapy‐induced nausea and vomiting. Other studies on discontinuation of PIMs in relation to time before death are scarce and report similar findings on use and discontinuation of lipid‐modifying agents,34, 35, 36 proton pump inhibitors (PPIs),36, 37 and neuroleptic antipsychotics.38 PPIs and neuroleptic antipsychotics are considered as problematic PIMs and candidates for deprescribing. Recently, clinical practice deprescribing guidelines, including an algorithm to guide deprescribing, were developed for both groups.39, 40 The high prevalence of initiation of both groups in our study raises questions about the dissemination and implementation of these guidelines in clinical practice. Lipid‐modifying agents are one of the few therapeutic groups of medication that are generally considered to be futile at the end of life because these medications have no short‐term benefit and no additional value for symptom relief. Clinical trial evidence has shown that these medications can be safely and effectively discontinued.16

For only one fifth of the population 75 and older was at least one PIM discontinued close to death, and for nearly half of the population no discontinuation was observed. This is consistent with Barcelo et al. who found that a large number of elderly patients with limited life expectancy continue to receive inappropriate medications.34 Clearly, there is no culture of discontinuation of PIMs at the end of life in Belgium. Apparently, many barriers to discontinuation or deprescribing exist. Overcoming these barriers is crucial to enable embedding of deprescribing in routine prescribing patterns. In order to be successfully implemented, all interventions to support physicians to engage in deprescribing should take these barriers into account.

Concordant with Chang et al., this study demonstrated that hospitalization is associated with discontinuation of PIMs.41 In Belgium, older adults are preferably hospitalized on a geriatric ward and treated by a multidisciplinary team including a geriatrician and healthcare professionals specialized in care for geriatric patients. Moreover, if hospitalized on another ward, the treating physician is encouraged to consult a geriatric support team—including a geriatrician—for patients with a positive geriatric risk profile, indicating increased frailty.42 Research has demonstrated that geriatricians prescribe fewer PIMs compared to other clinicians.33 Possibly, geriatricians are more aware of existing criteria and tools to identify PIMs and to support deprescribing and clinical practice deprescribing guidelines due to the need to change care goals and treatment targets in severely ill or frail older patients with limited life expectancy. Moreover, multidisciplinary collaboration, which is more prominent in hospital or in a nursing home compared to at home, in itself may lead to an increased attention for multidisciplinary medication review and discontinuation of PIMs, and partially explain the association we found between discontinuation of PIMs and hospitalization, and living in a nursing home.

Living in a nursing home was associated with decreased PIM use. This is consistent with Morin et al., who found a 15% reduction in the likelihood of receiving inadequate medications during the last month of life in institutionalized older adults with dementia.14 In Belgium, extensive home care facilities are available. Thus, nursing homes provide care for older adults with multimorbidity, severe functional impairment and increasing care needs that cannot be met in any other way. Ivanova et al. found that medication use in general in residents with dementia—who represented 46% of the population they studied at follow‐up—decreased between nursing home admission and follow‐up after two years in Flanders, the Dutch‐speaking part of Belgium.43 The high prevalence of dementia within the nursing home population may partly explain the association between living in a nursing home and discontinuation of PIMs. Another possible explanation is that the limited life expectancy after nursing home admission44 may lead to different, more cautious, patterns of prescribing and discontinuation of PIMs.

Chronic medication use was higher in the group for whom at least one PIM was continued. As the number of prescribed medications was found to be the main driver of PIM use in earlier studies,32, 33, 45 these people were likely to use more PIMs.

5.4. Implications for clinical practice and further research

More guidance on deprescribing in the context of limited life expectancy is needed in order to prevent unnecessary harm caused by PIMs at the end of life, taking into account prognostic uncertainty. Physicians urgently need practical evidence‐based guidelines and implementation plans, lists of candidate medications for deprescribing, training in how to initiate deprescribing and a better consensus on what inappropriate medication is. Furthermore, adaptation of existing international deprescribing guidelines to the context of limited life expectancy, in combination with a more realistic estimation of prognosis or prediction of death, is crucial to optimize medication use in this situation.

COMPETING INTERESTS

There are no competing interests to declare.

Paque K, De Schreye R, Elseviers M, et al. Discontinuation of medications at the end of life: A population study in Belgium, based on linked administrative databases. Br J Clin Pharmacol. 2019;85:827–837. 10.1111/bcp.13874