International changes in end-of-life practices over time: a systematic review

Background End-of-life policies are hotly debated in many countries, with international evidence frequently used to support or oppose legal reforms. Existing reviews are limited by their focus on specific practices or selected jurisdictions. The objective is to review international time trends in end-of-life practices. Methods We conducted a systematic review of empirical studies on medical end-of-life practices, including treatment withdrawal, the use of drugs for symptom management, and the intentional use of lethal drugs. A search strategy was conducted in MEDLINE, EMBASE, Web of Science, Sociological Abstracts, PAIS International, Worldwide Political Science Abstracts, International Bibliography of the Social Sciences and CINAHL. We included studies that described physicians’ actual practices and estimated annual frequency at the jurisdictional level. End-of-life practice frequencies were analyzed for variations over time, using logit regression. Results Among 8183 references, 39 jurisdiction-wide surveys conducted between 1990 and 2010 were identified. Of those, 22 surveys used sufficiently similar research methods to allow further statistical analysis. Significant differences were found across surveys in the frequency of treatment withdrawal, use of opiates or sedatives and the intentional use of lethal drugs (X2 > 1000, p < 0.001 for all). Regression analyses showed increased use of opiates and sedatives over time (p < 0.001), which could reflect more intense symptom management at the end of life, or increase in these drugs to intentionally cause patients’ death. Conclusion The use of opiates and sedatives appears to have significantly increased over time between 1990 and 2010. Better distinction between practices with different legal status is required to properly interpret the policy significance of these changes. Research on the effects of public policies should take a comprehensive look at trends in end-of-life practice patterns and their associations with policy changes.


Background
End-of-life policies are hotly debated in many countries, with international evidence used to support or oppose legal reforms [1][2][3]. For example, recent policy proposals on the legalization of "medical aid in dying", "physicianassisted suicide" and "euthanasia" in Canada, the United Kingdom, and France have made frequent references to empirical research from other countries [4,5].
A scoping review highlighted the need to take a comprehensive look at international end-of-life practice variations, because existing reviews tended to focus on specific practices or selected countries [6]. More specifically, trends in patterns of "euthanasia" and "assisted suicide" are most frequently discussed, with less attention being put on treatment withdrawal and the use of drugs for symptom management [7][8][9][10][11]. This could be problematic as it limits the ability to study the effects of public policies on the full range of end-of-life practices.
Also, changes in a single jurisdiction might reflect global trends in end-of-life practices, rather than domestic patterns.

Objective
The goal of this research was to systematically review international time trends in end-of-life practices.

Design
We conducted a systematic review of empirical studies on medical end-of-life practices (see Fig. 1 for the flowchart).

Definitions and classification of end-of-life practices
Definitions and labelling of medical end-of-life practices vary and no consensus exists on terminology at the international level. For example, definitions of Fig. 1 Flowchart of study inclusion and exclusion "euthanasia" have evolved over time and across jurisdictions [12,13]. For these reasons, we developed a descriptive classification of medical end-of-life practices. The classification is based on observable medical behaviors that can be studied empirically, and distinguishes practices with different legal status across jurisdictions, including: 1. Withdrawing or withholding of treatments that have the potential to prolong life; 2. Use of drugs for symptom management even if an unintended side-effect may be to shorten life; 3. Intentional use of a lethal drug.
The classification further distinguishes whether endof-life practices are carried out: a) with a voluntary and informed request made by the patient; b) with a voluntary advanced directive made by a previously competent patient, c) with a substitute request by the proxy decision-maker of an incompetent patient; or d) without a patient or substitute request. We also distinguish whether lethal drugs are administered by patients or by someone else.
Testing of this classification in a scoping review allowed for the translation of different end-of-life practice definitions into comparable categories, despite variations in labeling and definitions used in the original studies [6].

Inclusion criteria
To be included, articles written in English or French needed to: 1) report on physicians' actual end-of-life practices (rather than their opinions and attitudes); 2) include data on treatment withholding/withdrawal, use of drugs for symptom management, and the intentional use of lethal drugs; 3) allow estimates of annual practice frequency at the jurisdictional level. We excluded articles limited to specific populations (e.g. neonates) or practice settings (e.g. nursing homes). We also excluded articles using physicians' self-reporting of end-of-life practices to public regulatory authorities (e.g. euthanasia review committees) as this data source would not allow comparisons across jurisdictions with different legislative frameworks and would not provide information on some categories of practices included in our classification (e.g. intentional use of lethal drugs without patient request).

Search strategy
The literature databases searched were: MEDLINE, Embase, PsycInfo, Web of Science (SCI, SSCI), Sociological Abstracts, EMBASE, PAIS International, Worldwide Political Science Abstracts, International Bibliography of the Social Sciences and CINAHL. Search terms are listed in Appendix 1. The literature was searched in April 2014 and were pooled with articles identified in a previous scoping review that also included grey literature and online publications [6].

Data screening and extraction
Two reviewers separately screened each article and any discrepancies regarding the inclusion or exclusion of articles were resolved in team meetings. The information and statistics on these articles were extracted by one of two assistants and two review authors, verified by one of the review authors (GG or YSC), and entered in a FileMaker Pro v. 13 database developed with an information technology specialist. The items extracted from articles included year of publication, authors, title, abstract, source database, the types of end-of-life practice studied, funding sources, types of study (qualitative, quantitative or mixed), characteristics of study participants, and affiliations of first authors. When the same survey was reported in multiple articles, an index article was used to extract information on data collection methods (including duration, population groups and region of study), response rates, selective reporting for outcomes, details in statistical procedures, total numbers of deaths and summary percentages of end-of-life practices in each jurisdiction.

Comparability of study methods and risk of bias
The objectives for assessing the risk of bias were to 1) identify the risk of bias in estimated frequencies; 2) understand the heterogeneity of study methods to identify a subset of comparable studies using similar reference populations and outcome measures. We used quality appraisal criteria designed for systematic reviews [14,15] and observational studies [16,17]. The assessment criteria were: background information (population coverage, number of studied deaths, total number of deaths in the jurisdiction) [18]; basic survey information (article types, study design, author affiliations) [16]; design-specific assessment (statistical, conflict of interest, response rate, frequency of measurement, sample sizes, methods to adjust for selection bias) [19][20][21]; and practice-related assessment (methods for measuring outcome variables and selective outcome reporting) [22]. After reviewing the included surveys, the comparability of study design and statistical methods were assessed in the team meetings. Surveys using comparable methods (population characteristics, questionnaire types, statistical methods, and reporting methods) were retained for regression analysis.

Data analysis
The main outcome measure was the annual frequency of end-of-life practices as a percentage (%) of all deaths [7]. We further collected the 95 % confidence intervals (CIs) of the weighted frequencies or the nominators and denominators of the reported frequencies.
The extracted data were analyzed with R programming languange and RStudio (version 0.98.1091, RStudio Inc.). The surveys that adopted multiple-stage sampling usually provided the mean percentages of end-of-life practice with 95 % CIs. The comparisons of the percentages of end-of-life practices across jurisdictions were carried out using Chi-squared tests. If Chi-squared tests were not appropriate, simulation was used [23].
The relationships between the frequencies of endof-life practices and time (year of data collection) were assumed to be linear and were analyzed with logit regression. The null hypotheses were that there were no associations between the frequencies and time (regression coefficients of time equal to zero). Jurisdictions or survey types that were found to be significantly associated with different levels of endof-life practice frequencies [7] were adjusted in models that included all eligible surveys. Due to the variability originating from stratified sampling and adjustment for non-response, the 95 % CIs of reported frequencies were taken into account using simulations. The mean percentages were used to draw the regression lines in the graphs. The magnitudes of change in end-of-life practices from all included surveys were estimated with the differences in the predicted frequencies of the Dutch death certificate surveys from 1990 (or 2001 for use of sedatives for symptom management) to 2010 based on the regression models that included all surveys, since the Dutch surveys were the only ones to have been implemented throughout the period. The Dutch and Belgian surveys were also analyzed with separate regression models.

Results
Among 8183 references, 39 jurisdiction-wide surveys were identified. Of those, 22 surveys used sufficiently similar research methods to allow further statistical analysis and comparisons (Appendix 2). Response rates ranged from 40.0 to 84.0 % (median: 65.9 %). The 22 included surveys adopted or modified questionnaires that were developed in the Netherlands in the early 1990s [24] and used similar definitions of end-of-life practices, survey methods, and statistical analysis. Two main sampling strategies were used to identify responding physicians: random sampling of death certificates, and the use of physician registries. In these surveys, physicians were either interviewed individually or responded to selfadministered questionnaires. Figures 2, 3, 4, 5, 6 and 7 present end-of-life practice frequencies by year. The original studies did not allow to fully distinguishing end-of-life practices according to their legal status, as per our descriptive classification (Table 1). For example, original studies did not distinguish treatment withdrawal with or without voluntary patient request. Also, the use of opiates and sedatives did not distinguish whether these drugs were adjusted to symptom management or used with the intention to cause death.
Treatment withholding/withdrawal and the use of drugs for symptom management were the most frequent end-of-life practices. The intentional use of lethal drugs was less frequent, with frequency consistently below 5 % of all deaths.
The total predicted percentages of death in which physicians reported having made an end-of-life decision increased from 38.0 to 64.9 % between 1990 and 2010. Significant changes were found in the frequency of all end-of-life practices over time (p < 0.001 for all). The magnitude of change in the predicted frequencies of end-of-life practices was greater for the use of opioids (+15.27 % in annual practice frequency between 1990 to 2010) and for the use of sedatives with possible life-shortening effects (+6.78 %). Changes in frequencies were smaller for treatment withholding or withdrawal (−0.73 %), and for the intentional use of lethal drugs self-administered by patients (−0.17 %), administered by professionals with patient request (+0.71 %), or administered by professionals without patient request (−0.65 %). Changes in opioid use were significantly correlated with changes in: the use of sedatives, in the intentional use of lethal drugs administered by patients, and in the intentional use of lethal drugs without patient request (p < 0.05). Time trends estimated from Dutch or Belgian data showed the same directions of change as those predicted from all jurisdictions, except for withholding or withdrawal of treatment (p < 0.001).

Key findings
To the best of our knowledge, this is the first systematic review to take a comprehensive look at international variations in end-of-life practices over time from 1990 to 2010. This review showed an increase in the frequency of cases where physicians reported having made a decision that may have influenced the timing of death, which is consistent with previous findings [11]. A unique contribution of this review is its documentation of a significant increase in the use of opiates and sedatives over time, as opposed to other end-of-life practices.
The way opiate and sedative use is categorized in the original studies makes it difficult to interpret these time trends, because study questionnaires do not clearly distinguish the use of opiates and sedatives with or without the intention to hasten death [24]. For example, practices labeled in original studies as "intensification of symptom alleviation" include the use of drugs with a partial intention to hasten death [25][26][27]. Accordingly, the observed increase in the use of opiates and sedatives could mean that physicians are becoming more prone to use these drugs for symptom management. It may also reflect increasingly using opiates and sedatives with the intention of hastening patients' death among physicians. The observation that the use of opiates with possible lifeshortening effects is negatively correlated with the intentional use of lethal drugs could support the hypothesis of a "substitution effect" between different end-of-life practices. This suggests that drugs clearly associated with an intention to hasten death (e.g. neuromuscular blockers) are being replaced by opiates and sedatives in situations where the use of lethal drug is more difficult to justify in legal terms. Alternatively, these trends could simply reflect more professional willingness and patients' expectations to treat pain and symptoms at the end-of-life. Use of these drugs could have become more strongly embedded in clinical and social practices as more information becomes available about the beneficial use of opiates and sedatives for symptom management [28].

Policy and research implications
This review highlights the importance of exploring the potential effects of public policies on all end-oflife practices, rather than focusing exclusively on those targeted in policy documents. For example, a number of studies have explored the relationship between "euthanasia" legalization in the Netherlands and Belgium and the use of lethal drugs with the explicit intention to hasten death [29,30]. However, the review shows that the magnitude of predicted changes in intentional use of lethal drugs is much less than the use of opioids or sedatives with possible life-shortening effects. Focusing on small changes in the frequencies of intentional use of     This review also highlights the need for better articulation between end-of-life research and policy. The current confusion in empirical research between the use of opiates and sedatives with or without the intention to hasten death makes it difficult to interpret changing patterns in the use of these drugs because practices with different legal status are grouped together [6]. Better alignment between end-of-life practice classification in policy documents and empirical research would facilitate testing of policy-relevant hypotheses about the potential impacts of different policies on end-of-life practices. Our findings also underscore the importance of international research collaborations to harmonize study methods. Such collaborations have emerged in Europe (e.g. the EURELD consortium [31]), but are less developed in North America and elsewhere, which limits the potential for international comparisons.

Strengths and limitations
A strength of this review is the exhaustive efforts made to synthesize international evidence on a range of end-of-life practices. This approach offers a complementary perspective to previous reviews that focused on a single practice or a specific jurisdiction [7,10,29,32]. The use of a descriptive classification of end-of-life practices is a novel approach to systematically review evidence on end-of-life practice, and helped deal with international variations in labeling and definitions. This descriptive classification also clarifies areas of potential confusion when practices with different legal status were grouped together in original studies. Variations due to sampling and weighting procedures were also taken into account in the statistical analysis of end-of-life practice frequency. This is an improvement upon previous reviews that did not conduct statistical tests at all, or neglected the sampling variability of reported frequencies [7].
There are several limitations in this review. We only search for studies published in English and French. The increase in opioid use may result from both the intent to manage symptoms and to hasten death and we do not have evidence to assess this relationship. We found that comparative international evidence is limited by differences in study methods, end-of-life practice definitions, data collection approaches, sampling strategies, response rates, and target patient and professional groups. A limited number of surveys used similar methods allowing comparisons across jurisdictions and over time. Nonetheless, some degree of heterogeneity remains in the included surveys. For example, the two United Kingdom studies included surveyed different physician groups and used different versions of the same questionnaire [33,34]. Other European surveys also excluded children under one years old [31] or drew their sample from a specific region (e.g. Flanders in Belgium) [35]. As a result, some of the observed international variations may still reflect differences in study methods rather than actual differences in practice frequency. Also, only a limited number of European jurisdictions have conducted repeated surveys over time. Accordingly, regression models are largely driven by data from the Netherlands and Belgium, and observed time trends may be less generalizable to other jurisdictions. The lack of studies in other jurisdictions, such as the United States and Canada may also limit the generalizability of ther results.

Conclusions
The use of opiates and sedatives with possible lifeshortening effects appears to have significantly increased over time from 1990 to 2010. Treatment withholding/ withdrawal and the use of drugs for symptom management are the most frequent end-of-life practices. The intentional use of lethal drugs is less frequent, with frequency consistently below 5 % of all deaths. Better distinction between practices with different legal status is required to properly interpret the policy significance of these changes. Research on the effects of public policies should take a comprehensive look at changing the end-of-life practice patterns, rather than focus on a limited range of practices.  Witholding or withdrawal of treatment with the potential to prolong life « Non-treatment decisions »: the withholding or withdrawal of treatment in situations where the treatment would probably have prolonged life [24]; withhold or withdraw medical treatment while taking into account the possibility or certainty that this would hasten the patient's death or with the explicit intention of hastening the patient's death [31]; withheld or withdrawn medical treatment while taking into account the possible hastening of death (only for cases in which there was no single most explicit intention, the administration of drugs prevailed over the withholding or withdrawing of treatment) [36]. Note : Inconsistently report whether drugs were administered with or without patient voluntary request.

3
Use of opiates with possible life-shortening effects « Alleviation of pain and symptoms » : alleviation of pain and symptoms with opioids in such dosages that the patient's life might have been shortened [24]; intensify the alleviation of pain and suffering while taking into account the possibility or certainty that this would hasten the patient's death or partly with the intention of hastening the patient's death [31]. Note : includes cases where opioids were used with the intention to hasten death. Inconsistently report whether drugs were administered with or without patient voluntary request.

4
Use of sedatives with possible life-shortening effects « Continuous deep sedation »: deeply and continuously sedated until death [36]; the patient receive drugs, such as barbiturates or benzodiazepines, to keep him/her continuously in deep sedation or coma until death [26]. Note : does not distinguish whether sedatives were used with or without the intention to hasten death. Inconsistently report whether drugs were administered with or without patient voluntary request.

5
Intentional use of lethal drugs self-administered by patients « Assisted suicide »: the prescription or supply of drugs by a physician with the explicit intention of shortening life, when the drug is administered by patients [24,31]. Note : does not include the use of drugs with the « partial » intention to cause death.

6
Intentional use of lethal drugs administered by professionals with patient request « Euthanasia »: the prescription, supply or administration of drugs with the explicit intention of shortening life, when the dru gis administered by professionals [24]; death as the result of the administration, supply, or prescription of drugs with the explicit intention of hastening the patient's death (administered by professionals) [31]. Note : does not include the use of drugs with the « partial » intention to cause death.

7
Intentional use of lethal drugs administered by professionals without patient request « Termination of life without a patient request »: the prescription, supply or administration of drugs with the explicit intention of shortening life (without patient explicit request) [24]; death as the result of the administration, supply, or prescription of drugs with the explicit intention of hastening the patient's death (without patient explicit request) [31]. Note : does not include the use of drugs with the « partial » intention to cause death.    All attending physicians of the sampled cases in strata two to five (See question 6) received a questionnaire.
1) All deaths that occurred in that period were assigned to one of five strata. When the cause of death clearly precluded end-of-life decisionmaking), cases were assigned to stratum one. These cases were retained in the sample, but no questionnaires were sent out to the physician. Cases were assigned to one of the other strata looking at the likelihood that an end-of-life decision had preceded death: when this decision was unlikely cause of death was allocated to stratum two, when this decision was possible to stratum three, and when this decision was more probable (e.g., cancer) to stratum four. Cases were assigned to stratum five when the physician had noted on the death certificate that they had actively ended the life of the patient.         [2] the term "exclusion" meant the surveys excluded from statistical analyses with other surveys because of the focus on specific populations or geographic regions; [3] the article listed in the table was the one that used for data extraction; [4] the methods to identify physicians included death certificates, physicians (registries or associations), neonate/infants (patients or death certificates), or other specific patient groups; [5] the numbers in the column might be the total numbers of deaths that the frequencies of end-of-life practices could be applied or total numbers of death studied; [6] differential response meant the response rates might differ across sampling strata or subgroups; [7] selective reporting meant that not all results of a single study were reported in the article