Physician-related determinants of medical end-of-life decisions – A mortality follow-back study in Switzerland

Background Medical end-of-life decisions (MELD) and shared decision-making are increasingly important issues for a majority of persons at the end of life. Little is known, however, about the impact of physician characteristics on these practices. We aimed at investigating whether MELDs depend on physician characteristics when controlling for patient characteristics and place of death. Methods and findings Using a random sample (N = 8,963) of all deaths aged 1 year or older registered in Switzerland between 7 August 2013 and 5 February 2014, questionnaires covering MELD details and physicians' demographics, life stance and medical formation were sent to certifying physicians. The response rate was 59.4% (N = 5,328). Determinants of MELDs were analyzed in binary and multinomial logistic regression models. MELDs discussed with the patient or relatives were a secondary outcome. A total of 3,391 non-sudden nor completely unexpected deaths were used, 83% of which were preceded by forgoing treatment(s) and/or intensified alleviation of pain/symptoms intending or taking into account shortening of life. International medical graduates reported forgoing treatment less often, either alone (RRR = 0.30; 95% CI: 0.21–0.41) or combined with the intensified alleviation of pain and symptoms (RRR = 0.44; 0.34–0.55). The latter was also more prevalent among physicians who graduated in 2000 or later (RRR = 1.60; 1.17–2.19). MELDs were generally less frequent among physicians with a religious affiliation. Shared-decision making was analyzed among 2,542 decedents. MELDs were discussed with patient or relatives less frequently when physicians graduated abroad (OR = 0.65, 95% CI: 0.50–0.87) and more frequently when physicians graduated more recently; physician's sex and religion had no impact. Conclusions Physicians' characteristics, including the country of medical education and time since graduation had a significant effect on the likelihood of an MELD and of shared decision-making. These findings call for additional efforts in physicians' education and training concerning end-of-life practices and improved communication skills.


Methods and findings
Using a random sample (N = 8,963) of all deaths aged 1 year or older registered in Switzerland between 7 August 2013 and 5 February 2014, questionnaires covering MELD details and physicians' demographics, life stance and medical formation were sent to certifying physicians. The response rate was 59.4% (N = 5,328). Determinants of MELDs were analyzed in binary and multinomial logistic regression models. MELDs discussed with the patient or relatives were a secondary outcome. A total of 3,391 non-sudden nor completely unexpected deaths were used, 83% of which were preceded by forgoing treatment(s) and/or intensified alleviation of pain/symptoms intending or taking into account shortening of life. International medical graduates reported forgoing treatment less often, either alone (RRR = 0.30; 95% CI: 0.21-0.41) or combined with the intensified alleviation of pain and symptoms (RRR = 0.44; 0.34-0.55). The latter was also more prevalent among physicians who graduated in 2000 or later (RRR = 1.60; 1. 17-2.19). MELDs were generally less frequent among physicians with a religious affiliation. Shared-decision making was analyzed among 2,542 decedents. MELDs were discussed with patient or relatives less frequently when physicians graduated abroad (OR = 0.65, 95% CI: 0.50-0.87) and more frequently when physicians graduated more recently; physician's sex and religion had no impact. PLOS  Introduction [19]. Further evidence for the potential impact of physician's characteristics on shared decision-making came from a systematic review that found that one of the biggest facilitators for shared-decision making was the motivation of health professionals and their view that shareddecision making would lead to better patient outcomes [20]. The lack of knowledge surrounding how physician's characteristics impact real decisionmaking at the end of life-both the types of decision made as the involvement of others in this decision-makes it hard to address potential inequalities in decision-making, and to ensure shared decision-making is a priority in all end-of-life situations. With this study we aim to assess the impact of physician-related determinants of MELDs and patient involvement in these decisions. There are two specific research questions. First: Do physician characteristics have an impact on the likelihood of an MELD when controlling for both patient and setting characteristics? Second, do physician characteristics-in case an MELD took place-have an impact on the likelihood of patient and/or relatives being involved in shared decision-making when controlling for both the patient and setting characteristics?

Data collection
We conducted a mortality follow-back study on a continuous random sample of death registrations in Switzerland between August 7, 2013 and February 5, 2014, from which we obtained from the Swiss Federal Statistical Office the address of the certifying physician. The sample represented 21.3% of deaths among those aged 1 year or older in the German-, 41.1% in the French-and 62.9% in the Italian-speaking regions of Switzerland [21]. In total, 8,963 questionnaires were mailed in a strictly anonymous setting to the certifying physicians, of which 5,328 (59.4%) were returned until June 11, 2014.

Measures
Using an internationally standardized questionnaire [5] (online English version: [22]), physicians were asked whether they had: (1) withheld or withdrawn a probably life-prolonging medical treatment taking into account the possibility of hastening the patient's death or explicitly intending to hasten the patient's death or not to prolong their life; (2) intensified the alleviation of pain and/or symptoms (APS) with drugs taking into account or partly intending hastening the patient's death; or (3) prescribed or administered a drug with the explicit intention of ending the patient's life. For all these cases it was also assessed whether the patient ever expressed a wish for hastening death or for applying all life-prolonging measures, and whether the MELD was discussed with the patient or other persons (relatives, other physicians, healthcare professionals, any other person). Continuous deep sedation was assessed separately, but no questions were asked about the decision-making process so it is not included in this paper.
The region was defined by the language of the death certificate and place of death was determined from the categories in the questionnaire (at home, retirement community, long-term care home, hospice or palliative unit, hospital, other place).
Patient characteristics were available from death certificates (sex, age, civil status, religious affiliation). Broad cause of death was assessed in the questionnaire, because the cause of death information was not yet available from this early version of the death certificates and-due to anonymization-could not be supplemented ex post. More details are given elsewhere [23].
The questionnaire also encompassed several questions about the attending physician, namely sex, year (before 1970, 1970-1984, 1985-1999, 2000 or later) and place of graduation (German-speaking Switzerland, French-speaking Switzerland, abroad), the number of deceased patients cared for in the preceding six months, palliative care education (yes/no), religion or life stance (catholic, protestant, other religion/philosophy, not religious) and importance of religion/life stance for making an MELD (very important, important, less important, not important).

Sample
We selected all patients who were permanent residents of Switzerland, did not die suddenly and completely unexpected or by assisted suicide, and had a first contact with the responding physician when still alive. Out of 5,328 returned questionnaires, 3,391 concerned deaths that fulfilled these criteria and had at least minimal information on place of death and important physician attributes (sex, place and year of graduation).
For the analysis of shared decision-making, deaths without preceding forgone treatment or APS (N = 619) or lacking all information regarding discussion as well as other expression of patient's preferences (N = 230) were excluded, leaving 2,542 decedents with forgone treatment and/or APS and information about shared-decision making.

Statistical analysis
An age-sex-region-specific weighting was applied to make the data representative of all deaths in the sample period. Weighted percentages were used to describe the data.
Multinomial logistic regression [24] was used to calculate relative risk ratios (RRR), 95% confidence intervals (CI) and P values of potential determinants of different MELD categories. Binary logistic regression was used to calculate odds ratios, 95% confidence intervals (CI) and P values of potential determinants of any MELD vs. no MELD as well as of patient and family involvement in MELDs. All calculations were performed using Stata (version 13.1, StataCorp) statistical package.

Ethics approval and consent to participate
The study was declared exempt from ethics review by the Zurich Cantonal Ethics Board (KEK-StV-Nr. 23/13). Participants were informed about the study in a cover letter. Questionnaire return was considered to imply consent to participate.
The estimated relative risk ratios (RRR) from a multinomial logistic regression model contrasting "forgoing alone", "APS alone" and "both measures combined" with "neither" as reference category are presented in Table 2, along with 95% confidence intervals. Patient's nationality and religion were dropped in the final model, because they did not contribute information.
Compared to physicians that graduated from a Swiss university, those who graduated from abroad reported substantially less often forgoing treatment alone (RRR: 0.30; 95% CI: 0.21-0.41) or combining forgoing treatment with APS (0.44; 0.34-0.55) while for APS alone there was only a marginal effect. Having graduated in 2000 or later had only an effect for the combination of forgoing and APS (1.60; 1.17-2. 19), but not for the two "alone" categories. Forgoing life-prolonging treatment alone or combined with APS was less prevalent when physicians reported Catholic as their religious affiliation (0.71 / 0.72; 0.52-0.98 / 0.56-0.92), whereas among Protestant physicians the lower prevalence was predominantly driven by APS. The lowest prevalence however was found for the combination of forgoing and APS among physicians reporting another religious affiliation or philosophy of life (0.51; 0.37-0.71). Male physicians tended to report MELDs less frequently than their female colleagues. In all MELD categories, divorced patients substantially less often experienced an MELD (RRRs between 0.56 and 0.65).
It is tempting to line-up the impact of corresponding physician and patient determinants like sex, age, religious affiliation and foreign origin. The results of a multiple binary logistic regression model (adjusted additionally for the cause of death, place of death and language region) show that physician-related determinants had generally a larger effect size than patient-related ones (Fig 1). Compared to physicians that graduated from a Swiss university, those who graduated abroad reported significantly less often forgoing treatment or APS, whereas nationality of the patient was irrelevant. This contrast between a substantial effect among physicians but only minor effects among decedents also applied to religious affiliation. Only sex showed some similarity: Male patients and even more so patients cared for by a male physician were less likely to die following forgoing treatment or APS.

Patient and family involvement in medical end-of-life decisions
Most MELDs (79%) were discussed with the patient directly or with their relatives ( Table 3).
Involvement of the patient was more frequent among cancer patients than on average (53% vs. 37%). In multiple binary logistic regression analysis (not differentiating whether an MELD was discussed directly with the patient or only with relatives), however, there was no substantial variation between the broad cause of death groups. Nevertheless, the type of MELD had a large influence: Compared to forgoing a life-prolonging treatment alone, shared decision- . Conversely, shared decision-making was significantly less likely when physicians graduated outside of Switzerland (0.65; 0.50-0.87) or when patients died elsewhere than in a hospital. Physician's sex and religion or philosophy had no influence on the frequency of discussing decisions with the patient or relatives and were dropped in the final model. Compared to married decedents, shared decision-making was less frequent when patients were single (0.32; 0.23-0.45) or divorced (0.41; 0.29-0.58), but not when they were widowed. Patient's sex, age, nationality and religious affiliation were irrelevant and were also dropped in the final model.
The other potential physician determinants (palliative care education, the importance of religion/life stance for making an MELD, number of deceased patients cared for) were tested Odds ratio Physician-related determinants of medical end-of-life decisions Physician-related determinants of medical end-of-life decisions but did not have a significant impact on either the likelihood of forgoing and/or APS or the involvement of patients and relatives in decision-making.

Discussion
Even when adjusting for patient and setting characteristics, physician's characteristics had a substantial impact on MELD practices and the prevalence of shared decision-making in our study population, suggesting a sizable potential for optimizing end-of-life care. Physician-related determinants of medical end-of-life decisions International medical graduates made substantially fewer MELDs than those who graduated from a Swiss university and when they made an MELD, they less often reported shared decision-making with the patient and/or their relatives. International medical graduates may be less familiar with shared decision-making and patient-centered care [25]. At the time of the survey, 29% of all physicians in Switzerland were international graduates, with almost 60% of them originating from Germany [26,27]. A study revealed that German physicians chose more life-prolonging interventions than their Swedish peers [28] and a scoping review stated that "German physicians were found to be more likely to exclude patients, patients' families and non-medical staff from the decision-making process" [4]. In another study, it was shown that where physicians were trained had a significant impact on how often they discussed forgoing treatment with patients [19]. These findings support the notion that physicians' medical education may have a long-lasting impact on their attitudes towards care and decision-making. Physicians who graduated more recently made more MELDs and discussed these more often with patients and relatives than their colleagues who graduated before 1985. Similarly, a Belgian study found in the late 1990s that MELDs were significantly less frequent among patients treated by GPs aged 55 years and older [29]. This may also be due to previously lacking education in ethics: In Switzerland the first compulsory ethics classes started in 1995 only [30]. Ethics education has been described as supporting physicians' confidence regarding procedural end-of-life issues [31] and as being associated with a higher likelihood of applying a written do not resuscitate order [32]. Physicians inclined to apply "full code" had less often read about end-of-life care and had less interest in discussing MELDs than physicians more disposed to withdrawing life-sustaining therapies [32]. However, a 2002 study showed that ethics education was not associated with confidence in decisions to withdraw life support after an intensive care unit rotation and argued instead for experiential, case-based, patient-centred curricula for physicians-in-training [33].
Moreover and independently of age and place of graduation, physicians' religion mattered: physicians with a religious affiliation made fewer MELDs than those without. This is in line with a former study in Switzerland exploring attitudes regarding hypothetical patients and where religious believers tended to disagree more often with end-of-life decisions than other doctors [34]. On an international level, evidence for an impact of religious belief on the general incidence of MELDs is rather weak and controversial [12]. There is more evidence, however, for non-religious physicians being more inclined to make MELDs that may result in the hastening of death [16,29,35,36], and for a larger impact of religion for more drastic life-shortening acts [13]. Of note, the importance attached to religion when making an MELD had no influence on real patterns in our study and did not support the expectations of intrinsic religiosity of physicians playing a major role [12,34]. In contrast to our expectations, physician's sex had an only marginal influence on MELD incidence and no impact at all regarding the prevalence of shared decision-making. Nevertheless, future research should not discount the possibility of an effect of gender.
Except for the cause of death, civil status was the only patient characteristic with an impact on MELD prevalence. Fairly in line with other studies [6], MELDs were significantly less frequent among divorced patients. In Switzerland, this group accumulated more hospital days in their last year of life than other unmarried people, even when adjusting for the burden of disease and other sociodemographics [37], suggesting suboptimal health care provision. Discussion of MELDs was less frequent for divorced patients, too, as well as for single patients, however not for widowed patients. This is remarkable, since there is evidence that, irrespective of civil status, up to 95% of people had someone they would trust to make medical decisions for them [38]. However, strong social support and formal proxy decision-makers may be rarer among single and divorced patients, leading to fewer opportunities for communication with physicians.
Place of death had a substantial impact on patient involvement, with shared-decision making being more frequent in hospitals than in nursing homes or at home. This differs from the findings of an international study [39], which found that discussion with patients in most countries was more frequent at home than in institutions. Unsurprisingly, MELD category had a substantial impact on the prevalence of shared decision-making, with more than 80% involvement of patient/relatives when forgoing and APS are combined. In contrast, in the majority of cases of APS alone, physicians included neither patient nor relatives in the decision-making process, although they perceived their decision as a potentially life-shortening act.

Limitations
This study has several limitations. Most importantly, the observation unit were deaths and not physicians, i.e., several physicians filled in more than one questionnaire. Due to the anonymous nature of the survey, questionnaires stemming from the same physician could not be identified. Results therefore are not necessarily representative of Swiss physicians. Also, we cannot exclude the possibility of selection bias, since the response rate was, while considerable with respect to the setting, far from 100%. Non-response bias in shared decision-making is also an issue, since 243 questionnaires lacked all information regarding discussion as well as other expression of patient's preferences. However, non-response analysis revealed that the physician's characteristics of this group did not substantially differ from the study population, except an even higher proportion of physicians graduated abroad in the non-response group. The optimal phrasing of the questionnaire remains controversial [40]. This includes the way in which continuous deep sedation is queried, separately from actual MELDs and with less supporting information [41]. However, priority was given to maintain comparability with the international EURELD study [5] and the regular surveys in the Netherlands and Belgium. Finally, the questionnaire does not allow to test whether the actual MELD was in agreement to the preferences of patient and relatives.

Conclusions
The generally high prevalence of MELDs and shared decision-making in Switzerland support the notion that important goals like doctors' timely anticipation of end-of-life and departure from paternalistic medicine are largely accomplished. While there were few differences between patient groups in terms of MELDs or shared decision-making, divorced patients may be disadvantaged in the decision-making process and subsequent MELDs. Physicians should be proactive about engaging single and divorced patients in shared decision-making, possibly by identifying a proxy well in advance. The association between several physician's attributes and MELD practice points to the possibility of inequity in care and a substantial potential for improvement. The findings that older physicians and those graduated from abroad did not only make fewer MELDs but also if they made an MELD, they discussed it less often with patients, strongly call for additional efforts in residency training programs and physicians' vocational education in order to improve communication skills [42], preferably tailored to address local needs and context [43]. Communication has been called 'the cornerstone of good end-of-life care' [44]. An emphasis should also be given to strengthening physicians' motivation [16] and increasing awareness among patients and relatives that death is near [45], both being important elements in the process of improving end-of-life care.