Study design

Our survey comprised a questionnaire to record participants’ pPOC and pPOD. It consisted of closed-ended questions with occasional opportunities for free-text responses if the predefined answers were not suitable (for an English translation, cf. Supplementary Material). The items of the questionnaire were developed in a two-stage process. First, a group of physicians and nurses with different backgrounds reviewed the questionnaire developed by A.P. Prior to distribution, cognitive interviews were conducted with five PD patients to confirm face and content validity and to assess issues related to feasibility such as comprehensibility of the questions and acceptance [12]. The two cognitive interview techniques think-aloud and verbal probing [13] served to gain insights into the thought processes of the respondents. This provided valuable information regarding potential problems with questionnaire items, resulting in minor revisions being made to the questionnaire.

Study population and recruitment

The study was originally planned as a bi-center study in two different regions of Germany, but due to local restrictions caused by the SARS-CoV-19 pandemic, one center had to drop out. As a result, all participants were recruited from the Department of Neurology at the University Hospital of Marburg between June 2021 and September 2022.

In Germany, the care insurance system provides financial support to individuals in need of care, encompassing various care options such as home care, assisted living, and nursing home care. The objective of this comprehensive system is to ensure that individuals receive appropriate care that aligns with their preferences. Additionally, end-of-life care is facilitated by integrated health and social systems, with palliative care available in hospitals, hospices, and at home. The catchment area of the University Hospital of Marburg boasts a well-developed infrastructure that includes long-term care facilities, inpatient hospices, palliative care units, community nursing services, and specialist outpatient palliative care. However, there is currently no specialized neuro-palliative services in place.

To uphold ethical principles and maintain the reliability of the data collected, we excluded individuals with severe depression and cognitive impairment. These exclusions were deemed necessary to enhance the validity of our research findings. Detailed eligibility criteria for study participation are shown in Table 1. All patients who met the eligibility criteria were registered as potential study participants and asked about their willingness to participate in the study by a researcher who was not involved in the clinical care of patients.

Table 1

Eligibility criteria for study participation

Inclusion criteria	Exclusion criteria
Patients of any gender with a clinical diagnosis of PD according to MDS criteria [42]	Patients for whom the clinical team indicated that they were too weak and/or psychologically too stressed to participate in the study
Sufficient German language comprehension	Incapacitated persons, minors, and persons institutionalized by court or administrative order
Ability to give informed consent	Presence of life-threatening or severe psychiatric illnesses
	BDI-II>19 points
	MoCA<18 points
	Psychotic symptomsrequiring inpatient treatmentor affecting capacity to act
	Hearing or visual impairment of a severity that interferes with testing

Open in a separate window

Data collection

After patients’ informed consent, S.F. and J.K. conducted the interviews and clinical assessments. The final questionnaire was read aloud to all participants and the interviewer filled in the questionnaire manually according to the answers given but was also available for any queries or comprehension problems. Clinical assessments, including Hoehn & Yahr stage, Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) [14], Beck’s Depression Inventory II (BDI-II) [15], Montreal Cognitive Assessment (MoCA) [16], the age-adjusted Charlson Comorbidity Index (CCI) [17] and the Parkinson’s Disease Questionnaire (PDQ-39) [18] were administered during the interview, if not already done as part of the clinical routine. In instances where the clinical data were available, information was retrieved from the patient’s file.

Analyses

All statistical analyses were performed using R [version 4.12]. Prior to the study, we conducted a sample size calculation based on the calculation for odds ratios of 2.0 (equivalent to small effect sizes), which assumed a sample size of 94 and conservative estimates of small proportions of 0.25 with 95% confidence to provide a power of 80% [19, 20].

Descriptive analyses were performed on socio-demographic and clinical variables and data on end-of-life preferences. Due to the robust intercorrelation among measures of disease severity, principal component analysis (PCA) was employed in our study to mitigate the correlation between UPDRS Part III scores, Hoehn & Yahr stages, and Levodopa equivalent daily dosage (LEDD). The resulting principal component (PCA1) was subjected to further analysis as a proxy for disease severity. For an additional analysis of possible predictors of the dependent variables, i) preferred home care and ii) preferred home death, these were dichotomized and subjected to separate generalized linear models (GLMs). In addition, independent categorical variables were refactored where appropriate (e.g., German vs. non-German patients, etc.) resulting in the following predictors:

• 1)

  Gender

• 2)

  Age

• 3)

  Age at diagnosis

• 4)

  Disease duration

• 5)

  German nationality

• 6)

  Married

• 7)

  Religious/spiritual affiliation

• 8)

  Informal nursing support

• 9)

  Rurality

• 10)

  Professional education

• 11)

  Existence advance directive

• 12)

  Existence power of attorney

• 13)

  Palliative care knowledge

• 14)

  Hospice knowledge

• 15)

  Preferred place of care (for preferred home death), preferred place of death (for preferred home care)

• 16)

  Age-adjusted Charlson Comorbidity index

• 17)

  PDQ-39 score

• 18)

  BDI score

• 19)

  MoCA score

• 20)

  Disease severity.

All regression analyses were performed using the caret package [21]. We first estimated coefficients for the respective “full-model” with 20 predictors in a GLM using a binomial link function for the dichotomized dependent variables. Secondly, stepwise linear regression in both directions was used to identify important predictors of the two dependent variables out of all 20 possible independent variables. At each step, variables were subtracted based on the resulting Akaike Information Criterion (AIC) as a metric. To assess the model’s performance, 80% of the entire dataset was used (“training dataset”) and repeated ten-fold cross-validation ensured the robustness of performance results. The predictions of the two models were subsequently compared with the remaining 20% of data (“test dataset”) using accuracy, area under the curve (AUC) and logLoss as metrics. In a third approach, reduction of model complexity was tested using a regularized GLM with an ElasticNet approach [22]. This method has been extensively described, but briefly aims to identify coefficients minimising the sum of error squares by applying a penalty to these coefficients. ElasticNet has been suggested to be superior to other approaches in reducing the complexity of a model [23]. As before, the goodness of fit of the trained model (80% of the dataset) was tested on the remaining 20%, with robustness being sought through repeated 10-fold cross-validation. Values for accuracy, AUC, and logLoss served as metric to compare two reduced models to baseline. Finally, to obtain exploratory confidence intervals on the resulting coefficients, we chose a bootstrapping approach with 1000 resampling steps, using the identical methods as before. For the stepwise regression approach and the penalized model, the resulting coefficients were extracted to obtain their descriptive statistics.

The entire datasets and all performed analyses are available at https://github.com/dpedrosac/EOL_Parkinson.

Preferred place of care and preferred place of death

The pPOC at the end of life was the subject’s home for 60 participants (63.8%), a relative’s home for 2 (2.1%), a nursing home for 14 (14.9%), and an inpatient hospice for 5 (5.3%). A single participant (1.1%) expressed a preference for hospital care and none specifically for a palliative care unit. The POC was deemed not important for 9 participants (9.6%).

As for the pPOD, 41 participants (43.6%) specified their own household as their preferred location, while 4 (4.3%) referred to the home of a relative, 6 (6.4%) preferred a nursing home, and 8 preferred an inpatient hospice (8.5%). One participant stated a preference for dying in a hospital (1.1%), but no one expressed a desire to die in a palliative care unit. Meanwhile, 31 participants (33.0%) felt that the POD was not important to them.

Strengths and limitations

Our explorative research was conducted at a single center in Germany, which may somewhat limit the applicability to patients from other geographical regions. Future research with greater sample sizes and outside of Germany may confirm our results but also identify additional factors, including cultural and health system-specific characteristics of the pPOC and pPOD. Participants living in nursing homes and assisted living facilities were intentionally included to enrich diversity, and offer insights into the unique preferences within these care settings. This enhances real-world relevance, allows for a holistic exploration of influencing factors, and aligns with ethical considerations. However, under-representation may have occurred most likely due to eligibility criteria excluding many PwPD in advanced stages of disease, limiting generalizability. Moreover, our study has the limitations of a cross-sectional study; a longitudinal design could have provided additional information on the consistency of wishes. A mixed-methods approach would have allowed additional exploration of determinants, such as expectations, that are difficult to capture in a strictly quantitative research design. To better understand the decision-making processes in this context, further qualitative work is needed. Finally, our study lacks evidence on the congruence of pPOC and pPOD with the actual place of care and death, which could have provided deeper insights for service improvement and development.

The authors have no acknowledgments to report.