Various measures are implemented to ensure high data quality throughout the study. A key measure involves holding regular project meetings with the leading center and all staff involved in the study—including researchers and members of the palliative care teams—across all 16 study centers. Kick-off meetings are held both online and in person at the beginning of the study. Prior to data collection, all study team members receive standardized training on data collection and quality. This training, developed from years of experience with recruiting ICs in palliative care [34,35], lasts 90 minutes and is conducted by the leading center of this study. Trainings can be conducted online with staff from the same or different centers or on-site at the study centers. Written documentation of the training is provided. Additionally, individual or combined meetings are held at each study center to address specific conditions and requirements.

Additional mandatory and needs-based project meetings are scheduled throughout the data collection process to review insights, address uncertainties and challenges faced by the participating study centers, and share successful strategies. These different project meetings can help to increase involvement in the study and to minimize gatekeeping and insecurities in data collection [36].

To ensure secure and timely communication regarding the documentation of relevant information, multiple data transmission methods are available (eg, secured cloud systems and fax machines). All study processes and communication procedures have been verified and documented in a data protection plan and data protection impact assessment.

The study is overseen by a steering committee responsible for monitoring study progress and regularly evaluating data quality and procedural adherence. Meetings are convened 1 to 2 times per month, depending on the current phase of the study. The committee comprises CH, AD, PW, HS, and AU, with regular contributions from additional coauthors of this paper. It represents an interdisciplinary perspective, drawing on expertise in medicine, psychology, sociology, mathematics, and statistics. Any protocol modifications deemed necessary will be formally communicated to all relevant stakeholders, including study staff across participating centers, the study sponsor, and the responsible ethics committees. At the time of submission, the study protocol is version 1.3, dated March 31, 2025.

All physical study material (eg, signed informed consent forms) is securely stored in locked rooms at the study centers, while digitalized data are hosted on protected servers at the leading center. Data entry is centralized at the leading center, with regular duplicate entries performed on approximately 10% of baseline questionnaires and 2% of follow-up questionnaires, both randomly selected. Before data entry, study staff receive training on the standardized SPSS files and data entry protocols. To ensure consistency and accuracy, written documentation—including detailed codebooks and data entry guidelines for each questionnaire—is provided. Prior to data evaluation, the data will be cleaned in accordance with a set of predefined, written rules established a priori. This process includes range and plausibility checks as well as the review of handwritten entries. Access to the cleaned dataset will be restricted to researchers affiliated with the leading center and the designated scientific project leads at the participating study centers. Data will only be shared in pseudonymized or anonymized form to ensure participant confidentiality. Any additional exploratory analyses proposed by study centers will require prior review and approval by the leading center to ensure alignment with the overall study objectives and data governance protocols.

Given that this study is conducted in the vulnerable setting of SPC, several ethical considerations were made. The study follows the Methods of Researching End of Life Care by the UK Medical Research Council and National Institutes of Health Research in its planning and implementation [37]. Primary ethics approval was granted by the ethics committee of the Medical Association of Hamburg, Germany (2022‐100878-BO-ff), and, if required, secondary ethics approvals were obtained from the local committees of the other study centers.

During the data collection period, a regular telephone consultation hour with a psychologist in the leading center is offered for ICs to mitigate any potential study-induced burden. The consultation times and contact details are provided on each questionnaire.

A significant proportion of patients is expected to die within a few weeks after admission to SPC in Germany [38-40]. As patient death represents the end of this study, the information of patient death is documented in order to stop the dispatch of questionnaires. Additionally, if the patient dies in a facility not participating in the study, ICs are asked to report the death via the follow-up questionnaires, which include a sensitively phrased option for documenting the death. ICs are informed that no further questionnaires need to be completed. In the event of a patient’s death during the study, condolence cards are sent by the study team. Thank-you cards are sent if an IC drops out due to other reasons or concludes the study period.

To record ICs’ burden and needs, the baseline questionnaire and follow-up questionnaires consist of the following validated scales (German versions): DT of the National Comprehensive Cancer Network [41,42], Generalized Anxiety Disorder 7-item scale (GAD-7) [43,44], PHQ-9 [10,11,45], Short-Form Health Survey (SF-8) [46,47], Family Inventory of Needs (FIN) [48,49], and Oslo Social Support Scale (OSLO-3) [50,51]. While the FIN was specifically developed for assessing the needs of ICs of severely ill people, most of the questionnaires are generic (GAD-7, PHQ-9, and SF-8). The DT was originally developed for patients with cancer, but is also validated for the screening of ICs [9]. Previous studies have shown the feasibility of similar questionnaire surveys with ICs in SPC [34,52]. To align with the weekly assessment schedule, the recall periods of the GAD-7 and PHQ-9 were adapted from 2 weeks to 1 week. This adjustment was made to ensure consistency with the measurement intervals and to avoid overlap between consecutive assessments.

The instrument to be validated, the CAREPAL-8, is also part of each questionnaire. The CAREPAL-8 is a newly developed multidimensional 8-item screening instrument of IC burden in palliative care [12]. It was developed in a data-driven approach and consists of single items of the GAD-7, PHQ-9, SF-8, and FIN [12]. For the analysis, each item is dichotomized. By summarizing 4 items each, 2 indices are calculated: index 1 measuring psychopathological symptoms and quality of life (PSYQOL) and index 2 measuring the fulfillment of needs (NEEDS) [12]. By combining both indices, ICs are allocated to one of four groups with different risk profiles: (1) currently stable ICs (index 1=0 and index 2≤1), (2) ICs with unmet needs (index 1=0 and index 2≥2), (3) psychologically burdened ICs (index 1≥1 and index 2≤1), and (4) high-risk ICs (index 1≥1 and index 2≥2) [12]. For each group, the authors included clinical recommendations for further IC assessment and support [12]. Table 1 gives an overview of the included scales.

In addition to the scales mentioned, the baseline questionnaire includes questions on sociodemographic characteristics of both the IC and the patient, as well as potential associated factors. Table 2 gives an overview of the included items and scales. All baseline data were collected without planned missing values. The German baseline questionnaire is displayed in Multimedia Appendix 4.

To minimize the burden of the repeated data collection on the ICs, a design with planned missing data—a multiform design—is implemented for all follow-up questionnaires. As summarized by Little and Rhemtulla [31], this design reduces the number of items each participant answers by randomly splitting the total set of items across several forms. The participants are then randomly assigned to one of the forms, answering only the subset of items [31]. Previous research indicates that collecting data with this design leads to similar results compared to complete data designs and can improve validity by reducing participant burden and fatigue, lowering rates of unplanned missing data, and improving efficiency [31,32,61,62]. The multiform design used in this study is based on the work of Rhemtulla and Hancock [32]. A fixed block X consists of 14 items included in every follow-up questionnaire: current care situation of the patient (including an option to report death), all items from the CAREPAL-8, the 1-item analog scale of the DT, a free-text item about recent events that have influenced the burden and needs of the IC, as well as date and mode (with or without assistance) of questionnaire completion. The items from the following scales—GAD-7, PHQ-9, SF-8, OSLO-3, and FIN—are randomly assigned to 6 blocks (A-F), each containing 7‐8 items. The number of blocks was determined as a function of the time that ICs can be expected to spend completing the follow-up questionnaires (approximately 15‐20 minutes) and the number of items. Each follow-up questionnaire includes the fixed block X and 2 different blocks of A-F. Taking all combinations into account, this results in 15 different follow-up questionnaires. Table 3 gives an overview of the questionnaire composition.

The questionnaires are randomly distributed at each follow-up measurement point for every IC. The randomization is realized as follows: randomized sets of follow-up questionnaires are prepared for each study center, with each set containing an equal number of all 15 questionnaires (eg, each questionnaire is counted 5 times, resulting in sets of 5*15=75 questionnaires in total). If a study center collects data in more than 1 setting (eg, palliative care ward and specialist palliative home care team), each setting is provided with a separate set of randomized questionnaires to avoid imbalances. For every measurement point, each IC must receive a questionnaire from this predetermined set. Distributing a set of prepared questionnaires for multiple time points to one IC is not allowed to avoid imbalances that may occur if several questionnaires remain unanswered. The version (questionnaire number 1‐15) of the randomly allocated questionnaire is documented for each IC and measurement point. The implementation of the planned missing data design ensures that each follow-up questionnaire contains only half of the total items—29 to 30 instead of all 61—thereby relevantly reducing its length.

The cohort will be examined to identify groups with similar trajectories using growth mixture modeling (GMM) techniques [63]. More precisely, latent class growth analysis will be used to determine appropriate grouping structures among ICs. We will use the Bayesian information criterion as a relative measure of model performance and derive the most suitable number of trajectories from this. Factors associated with membership in these groups will be identified using a multinomial, multivariable regression model. Because the baseline assessment period was extended from the initially planned 3 days to 7 days, sensitivity analyses will be conducted using the 3-day subsample. As planned missing values are by design missing completely at random, we will use a multiple imputation approach to impute the missing data. Any additional unplanned missing values, potentially not missing completely at random, will be imputed in a further step if necessary. More precisely, first, we impute missing data at baseline on scale level after calculation of scales. In this step, only unplanned missing data exists. We only take this step for the data we use in the GMM. Second, we impute planned missing data on item level, set replaced unplanned missing data to missing data again, and then, calculate the scales and impute the unplanned missing data on scale level. The imputation model consists of all baseline and follow-up variables. Any missing values resulting from the patient’s death will be set to missing data again. We will use a joint modeling approach with a Weibull distribution of event times to account for missing data due to the patient’s death.

The data used to evaluate the psychometric properties of the CAREPAL-8 will be collected completely without planned missing values. To assess convergent validity, Pearson correlation analyses between index 1 and the scores of the GAD-7, PHQ-9, and both component scores of the SF-8, and between index 2 and the FIN will be conducted using baseline data. We expect high correlations of at least |r|=0.50. In our first version of the study protocol in 2022, we originally expected at least |r|=0.70. After further methodological considerations and before data analyses, we adjusted this value. According to the recommendations of Cohen [64], both values are interpreted as high. Sensitivity to change will be evaluated by using the patient’s care setting during the longitudinal observation as an external criterion, categorizing ICs in 2 groups: those who experienced at least 1 transition between outpatient and inpatient care (or vice versa) and those who did not transition. Clinically, it can be assumed that these transitions occur in response to significant changes in the patient’s health status or in the capacities of the ICs (eg, decompensation of home care due to IC overload) and are reflected in changes in the indices. We will evaluate whether a minimally important difference of 0.5 SD exists between these groups for each index. As an additional criterion, we will analyze the free-text item about recent events that have changed the burden and needs of the IC. The longitudinal data used in these analyses are part of the fixed block X, which is implemented in every questionnaire version. Supplementary analyses will assess the sensitivity and specificity of the instrument by determining cutoffs based on receiver operating characteristic analyses for both indices (index 1: PSYQOL and index 2: NEEDS). From this, we will analyze which cutoff for each index best predicts a change in care setting (transition vs no-transition group).

Statistical analyses will be conducted using the statistical software packages R (version 4.1.2 or newer; R Foundation for Statistical Computing) and SPSS Statistics (version 29 or newer; IBM Corp).

As the number of trajectory types is an outcome of this exploratory study, sample size calculation for GMMs is not feasible due to the uncertainty of key parameters. As a proxy, a mean comparison of the baseline variables between trajectory types is used. An IC is part of the analysis population if the baseline data and at least 1 follow-up questionnaire are valid. Various scenarios with 3 to 6 trajectory types were calculated.

Assuming, for reasons of clinical practicability, a highest number of 6 trajectory types and a small effect size (η²=0.025), we would need 85 ICs per group (total of 510 ICs) to detect this effect size with a power of 80% and a 2-sided type I error of 5%. This calculation assumes equal group sizes.

Sijbrandij et al [65] examined cases where estimation or performance issues arise when using GMMs. The authors concluded that no performance issues occur with sample sizes of 1000 individuals, while sample sizes of 300 individuals can present performance challenges, which can only be addressed by limiting variance parameters. Furthermore, we cite the study of Kim [66], which investigates GMMs in the sample size range targeted in this study. Additionally, we searched for studies using GMMs [67-72], which included sample sizes ranging from 129 to 6987 individuals. Our targeted sample size of 510 ICs is on the lower end of this range, but it is similar to another study conducted in the palliative care setting [71].

To calculate the required sample size for assessing convergent validity, we assume a Pearson correlation coefficient of 0.50 and a 2-sided 95% CI with a width of 0.20. Based on these parameters, a sample of 219 ICs is required. Therefore, by the end of the study, we aim to have baseline data from 510 ICs, along with at least 1 additional follow-up assessment, to demonstrate the described effects and apply the described models.

To apply the statistical models mentioned, baseline data plus at least 1 additional follow-up questionnaire from 510 ICs are required. Previous studies with ICs and similar inclusion and exclusion criteria have shown that baseline data could be collected from ICs of 30%‐40% of patients admitted to specialist inpatient palliative care [34,52]. Given the weekly follow-up assessments, a high degree of attrition must be anticipated in this study, primarily due to the rapid onset of patient death following admission to SPC or the study withdrawal of the IC. Consequently, we plan to recruit more ICs to account for this attrition. To facilitate study enrollment, we focus on the direct recruitment of ICs by trained study staff at the patient’s location of care. Flyers and other promotional materials for the study are available to support this process. To maintain the motivation of the ICs throughout the study, several measures are determined. First, all ICs are informed in a consistent and thorough manner about the study procedures, including the background, aims, and specifics related to the planned missing data design (eg, randomized order of items from different scales). Second, all study materials are carefully prepared and include contact information for any questions about the study as well as the psychological consultation hour. To facilitate the study processes for ICs, flexible options of receiving and transmitting the questionnaires are being offered: questionnaires can be distributed and returned at the participating study center or patient care location, sent by mail, or completed during a phone call. If questionnaires are not returned, study staff will follow up with ICs. The return rates of questionnaires as well as the overall data collection process are closely monitored so that potential difficulties and obstacles can be identified and addressed early on.

Other challenges may occur regarding the collaboration across 16 study centers in Germany. To ensure high data quality and comparability, a standardized and field-tested training for data collection and quality control is conducted at all study centers prior to data collection. Regular project meetings are held during the data collection period, and additional consultations are available at any time during the study process to address individual conditions. Due to the different conditions and resources across study centers, various ways for exchanging relevant study information between the centers and the leading center are implemented. For study staff without regular computer access (eg, a nurse in a specialist palliative home care team), relevant information can be documented on prepared paper forms and transmitted via fax or mail. Members of the study team who primarily work on computers (eg, scientific personnel) can use prepared Microsoft Excel sheets and share pseudonymized information via password-protected cloud systems. Quarterly newsletters are sent to inform study staff about past, present, and upcoming project activities, as well as recruitment progress and sample size. An expense allowance of US $319.2 (€280) is granted to each study center for every IC enrolled in the study with a valid baseline questionnaire.

With regard to the different SPC settings, recruitment strategies may differ. For example, the recruitment of ICs of patients who are cared for by multiprofessional SPC teams in hospitals might face challenges, for example, due to not meeting the ICs because of limited visiting hours in the hospital. We therefore developed different materials and recruitment procedures such as cover letters for ICs that can be deposited in the hospital room of the patient. If the patient transits between settings during the study period, data collection procedures might be adapted, for example, sending the questionnaire via mail instead of depositing the questionnaire at the location of care.

The multicenter design of the study, while essential for broad data collection and generalizability, presents potential challenges in ensuring transparent and equitable attribution of contributions across multiple institutions. To address this, authorship of resulting publications will be determined in accordance with the criteria established by the International Committee of Medical Journal Editors and the Contributor Role Taxonomy, ensuring that all contributions are appropriately recognized and documented.

The implementation of the planned missing data design involves several logistical and methodological challenges. Due to the random allocation of all items across 15 questionnaire versions, the preparation of the questionnaires required careful quality control to minimize the risk of errors. To address this, sufficient time was allocated for questionnaire preparation, and independent checks were performed by different members of the study team.

Particular attention is given to ensuring that the questionnaire layout is clear and does not increase confusion among ICs. Due to the random allocation of items, the thematic relationship between questions is not always apparent to participants, which may potentially lead to a perceived lack of coherence. To address this issue, we grouped items from the same questionnaires together and numbered items consecutively. All study-related information is presented in a nonintrusive manner, for example, by displaying the questionnaire version in a small font size in the lower right corner of each page.

All study staff are trained in the specific features of the planned missing data design and will provide ICs with appropriate instructions when distributing questionnaires. Randomization is centrally performed by the leading center as described in the Follow-Up Questionnaires (t1-t10) and Implementation of the Planned Missing Data Design section. As a result, each study center receives prerandomized questionnaire sets and is instructed to distribute the next available questionnaire within the set.

The design also requires specific documentation procedures. The questionnaire version administered to each IC at each measurement point is documented using structured paper- and electronic forms. Data entry is complex due to the existence of 15 questionnaire versions per measurement point, resulting in a structured and extensive data matrix. To ensure data quality, all personnel involved in data entry receive prior training, and double data entry procedures are implemented. This study is planned, conducted, and analyzed by an experienced multiprofessional research team, including experts in medical biometry.