Clinicians’ decisions about antibiotic treatment in advanced dementia: an international survey

Abstract

Objectives

In patients living with advanced dementia, the intensity of care during life-threatening infections remains controversial and marked by wide variation in practice. This international survey investigated physicians’ and physicians-in-training’s management choices for individuals with advanced dementia and the factors associated with those choices.

Design

Vignette-based survey.

Setting

Twelve countries across five continents.

Methods

We administered our vignette-based survey to medical students, residents and physicians. The survey elicited participants’ views on whether antibiotics should be administered to an elderly patient with advanced dementia and very poor quality of life, presenting with bacterial pneumonia. We explored factors associated with treatment choices using univariable analysis and multiple logistic regression models.

Results

Of the 785 participants (age, mean (SD): 31.1 (11.5) years), one-third (31.2%) resided in the Region of the Americas, 21.9% in Europe, 16.2% in the Eastern Mediterranean region and 17.1% in China. In the univariable analysis, choice to treat was associated with younger age, country/WHO region (African region highest overall, European region lowest overall), stage of medical training (medical student most inclined to treat) and absence of medical assistance in dying (MAiD) legislation. Multivariable analyses provided evidence that country was the variable most strongly associated with the choice to treat with antibiotics (Cameroon, China, Saudi Arabia highest; Norway, Switzerland, Spain lowest), with the presence of MAiD legislation also strongly associated (OR 0.35, 95% CI 0.23 to 0.51). Age (OR 0.82, 95% CI 0.66 to 1.00) and religiosity level (OR 0.93, 95% CI 0.87 to 0.99) showed weaker associations with treatment decisions.

Conclusion

Inclination to treat individuals with advanced dementia who develop pneumonia varies greatly between and within jurisdictions. Social factors (in particular country but also presence of MAiD legislation) proved the most prominent associations, with individual characteristics much less influential. These findings underscore the importance of contextual and cultural factors in value-sensitive clinical decisions.

Registration

We registered the protocol at Open Science Framework (osf.io/6kfbt).

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• In this international survey across 12 countries, clinicians’ decisions about antibiotic treatment in advanced dementia varied markedly by country, with contextual factors outweighing individual characteristics such as age, training stage or religiosity.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings highlight that value-sensitive clinical decisions are strongly embedded in cultural and legal contexts, with implications for clinician education, cross-border practice and guideline development.

Introduction

Consideration of patients’ values and preferences is a key component of optimal shared decision-making. Patients and clinicians have lived experiences that shape how they perceive the purpose of life and the meaning of death. Physicians must, with their patients and patients’ families, make over 100 decisions per day; occasionally, these decisions will have profound implications for patients’ lives.¹ Inevitably, decisions will be influenced not only by patients’ values and preferences but also by clinicians’ cultural environment, social relationships and prior experience.

A common situation in which values and preferences vary widely is the trade-off between quality and quantity of life, and the extent to which physicians should endeavour to prolong a life of poor quality. One context in which this dilemma frequently arises is the care of patients with advanced dementia. Cognitive decline in such patients typically progresses to the point that patients become entirely dependent and lack the capacity to make healthcare-related decisions. At this stage of the illness, clinicians must discuss with families critical decisions including do-not-resuscitate orders, using feeding tubes and hydration strategies and prescribing antibiotics for acute bacterial infections.^2–8

Although current guidance on the management of advanced dementia emphasises shared decision-making with relatives and caregivers,^9–11 families should base decisions on their understanding of what patients, when they had capacity, would have wanted if they developed advanced dementia. Both physicians and families may be uncertain regarding patient preferences.¹² Families may, in such situations, rely on guidance from physicians whose perspectives will influence how they conduct conversations, and sometimes whether they conduct them at all.^{13 14}

The use of antibiotics in patients with advanced dementia highlights the decisional challenges. Some argue that antimicrobial treatment is a potentially curative, life-prolonging intervention that risks extending suffering and indignity.^{15 16} Others contend that antibiotics may alleviate infection-related symptoms and reduce patient distress.¹⁷

Our past experience with questioning clinicians and medical students at educational events in a number of countries has suggested disparate views influenced both by local culture and stage of training. Uncertain whether rigorous study would confirm or refute these observations, we conducted a formal study in 12 countries across five continents investigating medical students’, residents’ and staff physicians’ views on whether antimicrobial treatment should be administered to an elderly patient with severe dementia, whose overall quality of life was extremely poor and who presented with community-acquired bacterial pneumonia.

Methods

This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.¹⁸ We registered the protocol at Open Science Framework (osf.io/6kfbt) and published the full protocol prior to study completion.¹⁹

Study design and setting

We conducted a vignette-based survey in 12 countries (Argentina, Brazil, Cameroon, Canada, China, India, Norway, Saudi Arabia, Spain, Switzerland, United Arab Emirates and USA) across five continents. With the aim of capturing substantial geographic and cultural variability, we purposively selected countries based on established collaborations within our research network. This approach facilitated recruitment from diverse cultural, religious, legal and healthcare settings while ensuring standardised study procedures through local investigators. At each site, a physician (MP, GSW, TA, SL, PA-C, FAN, RM, RA, AA, AFH, SK and AI) led site-specific study procedures. These investigators obtained local ethics approval, secured permission for survey administration, introduced the exercise to participants and answered questions from participants. Participants were unaware of the objectives of the study.

We administered paper-based or electronic questionnaires (REDCap v14.1.2), either in person or online, during educational sessions, conferences, rounds or via email to medical students, residents and physicians from November 2024 to November 2025. Our survey elicited participants’ views on whether antimicrobial treatment should be administered to an elderly patient with severe dementia who lacked decision-making capacity and had no available substitute decision maker, living in a nursing home, fully dependent on others for care, had very poor quality of life and who presented with community-acquired bacterial pneumonia. Online supplemental appendix 1 presents the full patient scenario and questionnaire.

To ensure accurate and consistent interpretation of the survey across countries, the study team developed, when deemed desirable by local investigators, surveys in each language; otherwise, respondents completed an English language survey. Translation proved desirable in two settings: Chinese in China and Portuguese in Brazil.

Participants

We enrolled medical students, residents and physicians who worked or studied in one of the 12 participating countries. Residents and physicians practised in one of the following specialties: internal medicine (including its subspecialties), intensive care, primary care/family medicine or emergency medicine. Participation in this survey was voluntary, and all respondents provided informed consent.

Questionnaire and variables

After reading the clinical scenario, participants chose between one of two options as the most appropriate conduct: (a) this patient should receive antibiotics or (b) this patient should not receive antibiotics. The vignette was intentionally designed to elicit participants’ normative judgements regarding goals of care, rather than to assess clinical knowledge or diagnostic reasoning. Respondents also had the opportunity to provide open-ended comments to explain their decisions. We collected the following demographic data from study participants: country, age, sex, gender, religion, religiosity, professional status ((1) medical student, (2) resident or (3) fully trained physician/fellow) and medical specialty. To capture religiosity levels while maintaining brevity, we collected one of the three domains (ie, intrinsic religiosity) from the 5-item Duke Religion Index (DUREL).²⁰ Intrinsic religiosity’s possible scores range from 3 to 15, with higher scores representing lower intrinsic religiosity. We also collected information on the legal availability of assisted dying in each country; throughout this manuscript, the variable called medical assistance in dying (MAiD) legislation encompasses any legally permitted form of assisted dying, including physician-assisted suicide and/or euthanasia, depending on the jurisdiction.

To assess the convergent validity of the single item regarding the desirability of antibiotic administration, participants answered the following question—“Should we always prioritize prolonging life, even when quality of life is very poor?”—with one of two response options: ‘(A) Yes’ or ‘(B) No’.

To verify the quality and interpretability of our survey, we pilot-tested in three countries (Canada, China and Brazil) with two staff physicians, one resident and two medical students at each site. Each participant provided feedback on the clarity, flow and relevance of the survey.

Statistical analysis

We conducted a descriptive analysis summarising categorical variables as counts and percentages, and continuous variables as mean (SD) for normally distributed data or median (quartile 1; quartile 3) for non-normally distributed data. We report the proportion of missing data for each variable.

Using the χ² test and the unpaired t-test or the Mann-Whitney U test, we conducted a univariable analysis to assess the association between predictors (age, gender, sex, country, WHO region, stage of training, religion, religiosity, confidence level and MAiD legislation) and antibiotic prescription. We also constructed a multivariable logistic regression model in which the dependent variable was the decision to prescribe antibiotics in the scenario presented (yes/no). The independent variables were entered into the model as a block. The following blocks were added: (1) demographic—age (years) and gender; (2) clinical—stage of medical training; and (3) cultural/contextual predictors—religion (Catholic, Protestant, Evangelical, Hindu, Buddhist, Taoist, Islamic, Jewish, no religion and other), religiosity, country where you work/study (Argentina, Brazil, Cameroon, Canada, China, India, Norway, Saudi Arabia, Spain, Switzerland, United Arab Emirates, USA) and presence/absence of a MAiD legislation in the country where the physicians practise. In model 1, we entered only the demographic block. In model 2, we added the clinical block. In model 3, we added cultural/contextual predictors such as religion, religiosity and country. We ultimately presented three variations of this final model: one in which each country was included separately (model 3A); another in which countries were grouped by WHO regions (model 3B); and third in which countries were dichotomised based on the presence of MAiD legislation (model 3C). To account for potential clustering by country and WHO region, we used cluster-robust SEs.

Multicollinearity was assessed through the examination of variance inflation factors (VIFs) for all independent variables, using a VIF ≥5 as an indicator of potential collinearity issues and linearity through Box-Tidwell test and visual inspection of scatterplots of the logit-transformed outcome versus each continuous predictor. Goodness of fit was assessed using adjusted R² and Akaike information criterion.

To measure the association between the validity question and the dependent variable, we used an OR ≥1.7 as our validity threshold. This cut-off point corresponds to a moderate association in most epidemiological research.²¹

We present our findings as OR with corresponding 95% CI and set a p value of 0.05 as the threshold for statistical significance. We used RStudio V.2024.12.1 to perform all statistical analyses.²²

Additional analysis

We conducted a sensitivity analysis using a multivariable hierarchical model, assuming a more in-depth nested data structure in which physicians and physicians-in-training were nested within their training stage (fixed effect), which is nested within countries (random effect). The same independent variables in model 3A were entered in this sensitivity analysis. In addition to prespecified analyses, we undertook post hoc, exploratory subgroup analyses focusing on respondents from China and Muslim-majority countries (United Arab Emirates and Saudi Arabia), as these groups had sufficient representation in our sample to permit meaningful within-group analyses. The objective was to examine within-subgroup variation in responses. We did not perform any imputation procedures to handle missing data.

To analyse open-ended comments, we conducted an inductive content analysis.²³ Two reviewers (JPL, JMS) independently generated open codes using Taguette, a qualitative data management tool, by identifying recurring ideas in these comments.²⁴ After coding, the reviewers met to discuss and resolve any discrepancies. Reviewers iteratively grouped similar codes and developed overarching categories.

Sample size

We used a sample size estimation for a binary logistic regression. For an OR as small as 1.9, assuming a binomial distribution, a level of significance of α=0.05, power (1−β)=80% and a multiple correlation of R²=0.2, we required a minimum of 511 participants. Online supplemental figure S1 presents a graph of OR by sample size. This computation was performed using G*Power 3.1.9.6.

Results

Participants

Table 1 presents participant characteristics. Of the 785 participants who responded to our survey (response rate: 71%), 541 (68.9%) preferred antibiotic treatment and 244 (31.1%) preferred to withhold antibiotics. The majority were women, with a mean age of 33.1. Fully trained physicians or fellows represented over 40% of the sample, residents and medical students each between 25% and 30%. Internal medicine and primary care proved the most frequently represented specialties.

Table 1. Characteristics of 785 participants.

Characteristic
Sex, n (%)
Female	486 (62.1)
Male	289 (36.9)
Prefer not to answer/missing	10 (1.3)
Gender, n (%)
Woman	478 (61)
Man	291 (37.2)
Two-spirit	4 (0.5)
Non-binary	3 (0.4)
Prefer not to answer/missing	9 (1.1)
Age (years)
Mean (±SD)	33.1 (±11.5)
Range	18–78
Country where you work or study
Region of the Americas, n (%)
United States	103 (13.1)
Canada	57 (7.3)
Argentina	43 (5.5)
Brazil	42 (5.4)
European region, n (%)
Switzerland	70 (8.9)
Spain	61 (7.8)
Norway	41 (5.2)
Eastern Mediterranean region, n (%)
United Arab Emirates	84 (10.7)
Saudi Arabia	43 (5.5)
Western Pacific region, n (%)
China	134 (17.1)
Southeast Asia region, n (%)
India	58 (7.4)
African region, n (%)
Cameroon	49 (6.2)
Religion, n (%)
No religion	302 (38.8)
Islamic	140 (18)
Catholic	134 (17.2)
Hindu	65 (8.3)
Protestant	43 (5.5)
Jewish	21 (2.7)
Evangelical	10 (1.3)
Buddhist	9 (1.2)
Christianism (other)*	8 (1.0)
Spiritism	6 (0.8)
Jainism	1 (0.1)
Taoist	1 (0.1)
Zoroastrianism	1 (0.1)
Prefer not to answer/missing	44 (5.6)
Religiosity (intrinsic religiosity DUREL index, 3–15)†
Median (Q1–Q3)	11 (6–15)
Stage of medical training, n (%)
Fully trained physician/fellow	343 (43.8)
Resident	209 (26.7)
Medical student	231 (29.5)
Missing	2 (0.3)
Specialty, n (%)
Internal medicine	328 (42.2)
Medical student	231 (29.7)
Family medicine/general practitioner	114 (14.7)
Intensive care medicine	30 (3.9)
Emergency medicine	23 (3.0)
Geriatrics	12 (1.5)
Neurology	12 (1.5)
Haematology–oncology	8 (1.0)
Cardiology	6 (0.8)
Palliative care	4 (0.5)
Rheumatology	3 (0.4)
Allergy and immunology	2 (0.3)
Endocrinology	2 (0.3)
Gastroenterology	1 (0.1)
Nephrology	1 (0.1)

^*Greek Orthodox (n=1), Christian Anglican (n=1), Christian Orthodox (n=3), Armenian Apostolic (n=1), Christian (unspecified) (n=1), Eastern Orthodox (n=1), Coptic Orthodox (n=1).

^†Higher scores indicate lower religiosity.

DUREL, Duke Religion Index; Q1, first quartile; Q3, third quartile.

Approximately one-third of the respondents worked or studied in the Region of the Americas, with approximately 20% in Europe and between 15% and 20% in the Eastern Mediterranean region and China (figure 1). Almost 40% of participants reported not having a religion. Among individuals reporting a religion, they most commonly identified as Muslim, Catholic or Hindu. Overall, religiosity had a median score of 11 (DUREL index ranging from 3 to 15; higher scores indicating lower religiosity).

Figure 1. World heat map of respondents by country. Colour intensity is proportional to the number of respondents, with lighter shades indicating higher numbers.

Questionnaire validity

The decision on whether to administer antibiotics in our clinical scenario showed a strong association with the validity question, “Should we always prioritize prolonging life, even when quality of life is very poor?”, with OR=7.78.

Predictors of antimicrobial treatment decision

In the univariable analysis, age, country/WHO region, stage of medical training, religion, religiosity level and the presence of MAiD legislation were associated with participants’ decisions to prescribe (or not) antibiotics. Medical students were more likely to choose antibiotic treatment (78.8%) than fully trained physicians/fellows (65.9%) and residents (62.7%) (p<0.001). Participants practising in countries with MAiD legislation were more likely to prefer withholding antibiotics (55.9%) than those in countries without MAiD legislation (23.3%) (p<0.001). Table 2 presents univariable analysis results in detail.

Table 2. Univariable analysis of antibiotic decision groups.

Characteristic	Treat with antibiotics (n=541)	Do not treat with antibiotics (n=244)	P value
Age, mean (SD)	32.04 (10.71)	35.5 (12.66)	<0.000
Gender, n (%)			0.60
Men	198 (68)	93 (32)
Women	330 (69)	148 (31)
Other*	6 (85.7)	1 (14.3)
Sex, n (%)			0.67
Male	196 (67.8)	93 (32.2)
Female	338 (69.5)	148 (30.5)
Country, n (%)			<0.001
Argentina	25 (58.1)	18 (41.9)
Brazil	32 (76.2)	10 (23.8)
Cameroon	48 (98)	1 (2)
Canada	29 (50.9)	28 (49.1)
China	125 (93.3)	9 (6.7)
India	44 (75.9)	14 (24.1)
Norway	4 (9.8)	37 (90.2)
Saudi Arabia	37 (86)	6 (14)
Spain	30 (49.2)	31 (50.8)
Switzerland	24 (34.3)	46 (65.7)
United Arab Emirates	67 (79.8)	17 (20.2)
USA	76 (73.8)	27 (26.2)
WHO region, n (%)			<0.001
Region of the Americas	162 (66.1)	83 (33.9)
European	58 (33.7)	114 (66.3)
Eastern Mediterranean	104 (81.9)	23 (18.1)
Western Pacific	125 (93.3)	9 (6.72)
Southeast Asia	44 (75.9)	14 (24.1)
African	48 (98)	1 (2)
Stage of training, n (%)			<0.001
Fully trained physician/fellow	226 (65.9)	117 (34.1)
Resident	131 (62.7)	78 (37.3)
Medical student	182 (78.8)	49 (21.2)
Religion, n (%)			0.006
No religion	191 (63.2)	111 (36.8)
Islamic	113 (80.7)	27 (19.3)
Catholic	83 (61.9)	51 (38.1)
Hindu	47 (72.3)	18 (27.7)
Protestant/Evangelical	38 (71.7)	15 (28.3)
Jewish	15 (71.4)	6 (28.6)
Other†	48 (75)	16 (25)
Religiosity, median (Q1–Q3)‡	10 (5–14)	13 (9–15)	<0.001
Confidence level, median (Q1–Q3)§	3 (3–4)	3 (3–4)	0.21
MAiD is legal in the country, n (%)¶			<0.001
Yes	83 (44.1)	105 (55.9)
No	458 (76.7)	139 (23.3)

P<0.05 values are shown in bold.

Categorical variables were compared using the χ2 test. Continuous variables were compared using the independent samples t-test or the Mann-Whitney U test, as appropriate.

Percentages are calculated within each category (row percentages).

^*Two-spirit and non-binary.

^†Buddhist (n=9), Christianism (other) (n=8), Spiritism (n=6), Jainism (n=1), Taoist (n=1), Zoroastrianism (n=1), prefer not to answer/missing (n=44).

^‡Intrinsic religiosity DUREL index ranges from 3 to 15; higher scores indicate lower religiosity.

^§1–4; higher scores indicate greater confidence.

^¶Spain, Canada and Switzerland are the countries that have MAiD legislation.

DUREL, Duke Religion Index; MAiD, medical assistance in dying; Q1, first quartile; Q3, third quartile.

Table 3 presents the results of the multivariable analyses. In model 1, we included only age and gender; age but not gender proved a significant predictor. In model 2, we added stage of medical training; age remained as a significant predictor, stage of training proved not significant (R²=0.025).

Table 3. Multivariable logistic regression results.

	Model 1		Model 2		Model 3A (clustering by country)		Model 3B (clustering by WHO region)		Model 3C
	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value
Age (per decade)	0.74 (0.66 to 0.90)	<0.001	0.74 (0.60 to 0.90)	0.004	0.90 (0.66 to 1.22)	0.58	0.90 (0.74 to 1.11)	0.43	0.82 (0.66 to 1.00)	0.06
Gender (ref=Man)
Woman	0.93 (0.67 to 1.28)	0.66	0.93 (0.67 to 1.29)	0.68	0.93 (0.66 to 1.30)	0.66	0.96 (0.68 to 1.35)	0.81	0.90 (0.64 to 1.28)	0.57
Other*	2.42 (0.39 to 46.73)	0.42	2.02 (0.32 to 39.32)	0.53	1.77 (0.11 to 27.19)	0.68	1.90 (0.09 to 38.91)	0.67	3.29 (0.48 to 66.96)	0.30
Stage of training (ref=Fully trained physician/fellow)
Resident			0.63 (0.40 to 0.99)	0.043	0.75 (0.45 to 1.24)	0.26	0.77 (0.54 to 1.10)	0.14	0.75 (0.46 to 1.21)	0.24
Medical student			1.11 (0.65 to 1.89)	0.69	1.05 (0.52 to 2.13)	0.89	1.06 (0.46 to 2.45)	0.90	1.23 (0.70 to 2.17)	0.47
Religion (ref=No religion)
Islamic					1.35 (0.50 to 3.63)	0.55	0.96 (0.29 to 3.18)	0.95	0.87 (0.42 to 1.83)	0.72
Catholic					0.69 (0.48 to 0.99)	0.043	0.83 (0.45 to 1.55)	0.57	0.69 (0.41 to 1.18)	0.17
Hindu					1.18 (0.36 to 3.87)	0.78	0.84 (0.34 to 2.10)	0.70	0.79 (0.40 to 1.59)	0.51
Protestant/Evangelical					1.61 (0.79 to 3.28)	0.27	1.05 (0.92 to 1.20)	0.44	0.92 (0.43 to 2.05)	0.84
Jewish					1.30 (0.73 to 2.31)	0.37	1.60 (1.22 to 2.10)	<0.001	1.46 (0.54 to 4.44)	0.47
Other†					1.41 (0.64 to 3.10)	0.40	1.16 (0.67 to 2.01)	0.60	1.19 (0.59 to 2.48)	0.64
Religiosity (per unit)‡					0.98 (0.93 to 1.02)	0.29	0.97 (0.96 to 0.98)	<0.001	0.93 (0.87 to 0.99)	0.015
MAiD is legal in the country (ref=No)§
Yes									0.35 (0.23 to 0.51)	<0.001

P<0.05 values are shown in bold.

^*Two-spirit and non-binary.

^†Buddhist (n=9), Christianism (other) (n=8), Spiritism (n=6), Jainism (n=1), Taoist (n=1), Zoroastrianism (n=1), prefer not to answer/missing (n=44).

^‡Intrinsic religiosity DUREL index, 3–15; higher scores indicate lower religiosity.

^§Spain, Canada and Switzerland are the countries that have MAiD legislation.

DUREL, Duke Religion Index; MAiD, medical assistance in dying; ref, reference.

Finally, in model 3, we analysed cultural/contextual variables. Religion and religiosity level were added to the analysis in all three variations (A, B and C). In model 3A, model fit improved substantially after accounting for clustering by country (R²=0.22) and showed Catholic religion to be a significant predictor, with lower odds of prescribing antibiotics. After clustering by WHO region, model 3B demonstrated modestly reduced model fit compared with model 3A (R²=0.19) while indicating Judaism and religiosity level as significant predictors. In model 3C, lower religiosity level and existence of MAiD legislation were both associated with a preference for withholding antibiotics (R²=0.1). Our analyses did not detect any violation of collinearity or linearity assumptions.

In a sensitivity analysis using a multivariable hierarchical model with country specified as a random effect and all other covariates treated as fixed effects, the intraclass correlation coefficient=0.32 indicated that a substantial proportion of the variability in treatment decision-making was attributable to country-level clustering. These results are consistent with those observed in model 3A (online supplemental table S1). Subgroup analysis of participants within China (online supplemental table S2) demonstrated that those who opted to withhold antibiotics were, on average, almost a decade younger than those who preferred antimicrobial treatment (mean (SD) 30.44 (8.74) vs 21.33 (4.15); p<0.000). The analyses including only Muslim-majority countries (United Arab Emirates and Saudi Arabia) also identified younger age as an important variable associated with withholding of antibiotics (mean (SD) 30.29 (8.21) vs 26.59 (5.89); p=0.19) (online supplemental table S3).

Inductive content analysis of open-ended comments

Of the 785 participants, 126 (16%) commented on their responses, 42 of whom opted to withhold antibiotics. A common (n=32) concern in this group was prolonging patients’ suffering. Comments included some that “the patient is dying; antibiotics would only prolong suffering” or that the patient has “no meaningful quality of life.” Others (n=10) focused on a diagnostic rather than a quantity-versus-quality-of-life perspective. For instance, two respondents wrote comments suggesting a “need to confirm the source of infection before deciding” and that “PE [pulmonary embolism] should be ruled out first.”

Over one-quarter (n=22) of the comments of the 84 participants who preferred antibiotics advocated antibiotics as part of comfort measures, arguing that “if the infection is causing distress, antibiotics may help with comfort.” Others (n=16) preferred to default to treatment: “Default to antibiotics until wishes are known.” Similar to the group of participants who withheld antibiotics, some (n=21) approached this clinical scenario with a diagnostic rather than a quantity-versus-quality-of-life perspective: “Diagnosis unclear; further labs or imaging are needed.”

Discussion

This cross-cultural survey conducted in 12 countries documented large differences both within and between jurisdictions in physicians’ and soon-to-be physicians’ management of patients with severe dementia with potentially life-threatening pneumonia. With respect to factors associated with the decision, our multivariable analyses suggested that the country in which participants work is the factor most strongly associated. Both our univariable and modelling analyses provide insight into the social and cultural environments that drive country influence. Presence of the option of ending one’s life with MAiD was strongly associated with willingness to withhold antibiotics in all analyses, and the dominant religious affiliation—if any—also appeared related to countries’ apparent influence on choices. Data suggested personal characteristics such as age and religiosity were generally less influential than contextual factors.

Our qualitative analysis provided important context for participants’ decisions. Respondents who preferred to withhold antibiotics most commonly cited concerns about prolonging suffering, whereas those favouring treatment frequently viewed antibiotics as a means of improving comfort or defaulted to treatment in the absence of known patient wishes. Notably, participants in both groups sometimes focused on diagnostic uncertainty rather than the intended ethical trade-off.

The greatest strength of our study is our success at recruiting an appreciable sample of physicians in our target population in 12 countries with wide variation in gross national product, religious practices and cultural attitudes. We achieved a high response rate in all our target populations, from medical students to practising physicians. Our analysis of respondents’ comments provides insights into the rationale for their choices. One limitation of this study is that we intended to explore physicians’ views on the ethical issues that arise in the management of patients with acute medical conditions and advanced dementia. However, a small number of respondents focused on diagnostic uncertainty and the need for further investigation. Had we anticipated this, we would have clarified that diagnostic evaluation was complete and the diagnosis of pneumonia was secure. Importantly, this response pattern occurred across both treatment groups and jurisdictions, making it unlikely to explain the substantial between-country differences observed in treatment preferences. In addition, our findings may be subject to selection bias because participation was voluntary and respondents may not reflect the broader clinician population from each included country.

With regard to prior work in this area, a vignette-based survey conducted in Quebec, Canada, with seniors, caregivers, nurses and physicians found that participants who find MAiD acceptable and with lower religiosity tend to favour withholding of antibiotics for patients with advanced dementia.²⁵ There is one striking discrepancy in findings: the Canadian sample in our survey shows approximately equal numbers between those who prefer withholding antibiotics (49.1%) and those who chose antibiotic treatment (50.9%), while the study conducted in Quebec reports that the majority of stakeholders find it ‘totally acceptable’ to withhold antimicrobial treatment.²⁵ The explanation may be survey location: findings are consistent with recently released data that show a disproportionate number of MAiD provisions in Canada occurred in Quebec (36.4% in a province in which approximately 22% of Canadians reside).²⁶ Most participants in our survey reside in Ontario.

Our findings contrast with a recent systematic review and meta-synthesis identifying psychological determinants of physician variation in end-of-life treatment intensity.²⁷ That review primarily synthesised qualitative and mixed-methods studies conducted within single jurisdictions, and highlighted individual factors such as professional identity, communication avoidance, decisional avoidance and personal values favouring life extension as contributors to aggressive care at the end of life. Our study suggests that cultural factors play a much stronger role than do physicians’ individual characteristics.

The finding that setting—in particular country—rather than individual characteristics is the strongest driver of clinicians’ ethical views on management of patients living with dementia while providing insight on the power of social factors raises the question of what it is about clinicians’ country that so powerfully influences their perspective. Dominant religion appears to be one factor. Saudi Arabia (86%) and United Arab Emirates (79.8%), two Islamic faith countries, have among the highest proportions choosing antibiotic administration, while India (75.9%), with a large Muslim minority, followed the same pattern. Muslim religion was not prominent in the regression analyses likely because of the relatively high number of Chinese participants (more than any other country, and several times the number in most other countries), the very high proportion of Chinese who would administer antibiotics (93.3%) and of respondents who identified as having no religion (92.5%).

Legislation allowing MAiD—in all cases passed in the last few years—had a strong association with preference for withholding antibiotics. Switzerland (65.7%), Spain (50.8%) and Canada (49.1%)—all countries with MAiD legislation and, aside from Norway, had the highest proportion not treating. Canada, which, along with the Netherlands, has the highest percentage of deaths through MAiD (approximately 5% in 2023) of any country,²⁶ contrasts with the USA. Despite similar cultures in many aspects, in Canada 49.1% chose no antibiotics, compared with 26.2% in the USA. Both the availability of MAiD and the choices of physicians in our survey identify cultures in which quality relative to duration of life is more highly valued. The most extreme results, Norway (over 90% no antibiotics) and Cameroon (98% choosing antibiotics), reflect powerful cultural features that our results do not explain. In Cameroon and India, health system factors—such as liberal antibiotic use and over-the-counter availability²⁸—may be associated with the increased preference for antibiotic treatment. Overall, these differences may also reflect broader country-level factors, such as prevailing norms regarding palliative care, clinician training and expectations around goals-of-care discussions, which could not be disentangled from legal contexts in this study.

In our exploratory analysis restricted to participants from China, those who opted to withhold antibiotics were, on average, nearly a decade younger than those who favoured antimicrobial treatment. This finding should be interpreted cautiously, as most respondents who chose to withhold antibiotics in this subgroup were medical students, and stage of training may partly explain the observed age difference. The difference may reflect a broader generational shift in Chinese society, economy and cultural values—from a long-standing emphasis on life preservation towards a greater prioritisation of quality of life and contemporary interpretations of life’s meaning.²⁹ While speculative, this interpretation aligns with the view that clinical decision-making is shaped by individual characteristics and evolving social and cultural contexts.

This study provides insights for physicians. In almost every country, whatever one’s own perspectives and beliefs, one will find colleagues with whom one works with different viewpoints. For physicians moving—definitively or temporarily—to another country, our results provide insight into the dominant approach to ethical issues in the management of patients with advanced dementia that they will encounter. Our findings may prompt changes to education in undergraduate and postgraduate medical training programmes. Such modifications may emphasise earlier integration of value-sensitive decision-making, goals-of-care discussions and their timing, and case-based teaching focused on advanced dementia.

Conclusion

Physicians’ and medical students’ choices regarding care of acutely ill patients with advanced dementia vary a great deal both within and between countries. Cultural factors appear to be highly influential in attitudes and perspectives. These results may prompt clinicians to reflect on the determinants of their attitudes, promote awareness and acceptance of differences and encourage more thoughtful approaches to the ethical care of patients with advanced dementia.

Data availability statement

Data are available upon reasonable request.