There were 24 study sites recruited by the research network of the Palliative Medicine Working Group (APM) of the DKG. In these cancer centers patients over 18 years with solid tumor entities were identified at the time of diagnosis of incurability (i.e., prior to initiation of palliative anti-cancer therapy and according to the definition of incurability as established in the former APM study (13)) via outpatient clinics, oncological wards or multidisciplinary tumor boards by treating physicians. Non-compliance and age being under 18 years were the only exclusion criteria. After study inclusion, the PI (Göttingen) assigned the patients to the two intervention arms (low-threshold screening versus comprehensive assessment) by using block randomization, stratified by center and tumor entity at a ratio of 1:1 ( Figure 1 ). Within this study, patient intervention took place only via conducting surveys. The patients were requested to complete simple screening surveys autonomously on regular follow-up visits to their disease.

A population of 504 patients was estimated to be recruited for this study including an expected dropout rate of 20%. For the primary endpoint in the SCREBEL study we expected a standard deviation of 17.26 based on the experience of the former APM study (12, 13). Thus, a non-parametric test on differences between groups regarding FACT-G with a two-sided significance level α = 5% yields a power of 80% if differences are at least 5 points, which represents a clinically relevant difference (18).

As the physicians were advised to include the results from the latest assessment into their treatment decision in order to improve the patient’s QoL we chose QoL as the primary endpoint. Quality of life was assessed using the following questionnaires:

The FACT-G (Functional Assessment of Cancer Therapy) (19) is a 27-item questionnaire designed to measure four domains of QoL in cancer patients: Physical, social, emotional, and functional well-being. The items are measured on a five-point Likert-scale from 0 (not at all) to 4 (very much). The score is the sum of all items and ranges from 0 to 108. The higher the score, the better the QoL.

The NCCN Distress Thermometer (National Comprehensive Cancer Network distress thermometer) (20) is a validated, widely used screening measure. The screening contains a single‐item visual numeric scale ranging from 0 (“no distress”) to 10 (“extreme distress”) to quantify the global level of distress experienced in the past week. A higher score indicates a higher distress and thus a lower QoL (21).

IPOS (Integrated Palliative Care Outcome Scale) (22) is a 10 question survey developed to measure palliative care needs of patients and their families. The questions address how limited the individual is due to the symptoms rather than the severity of the symptoms themselves. Of all questions, only the questions 2 (with again 10 subitems) and 3-9 contribute to the overall score, resulting in 17 contributing items. All items are measured on a five-point Likert scale (0 to 4). The IPOS is the sum of the 17 items mentioned above, thus ranging from zero to 68. A higher IPOS score indicates a lower QoL. Moreover, in the presence of at least two questions answered with “3” or at least three questions answered with “2”, further exploration and medical treatment is recommended.

PHQ-4 (Patient Health Questionnaire) (23) is a four item questionnaire addressing a patient’s psychosocial condition regarding anxiety and depression. The items are measured on a four-point Likert-scale from 0 (not at all) to 3 (nearly every day). The total score is the sum of all items and ranges from 0 to 12. A total score ≥ 3 for the first two questions indicate anxiety. A total score ≥ 3 for the last two questions suggests depression. The higher the score, the lower the QoL.

SCNS-SF-34-G (Supportive Care needs Survey – short form 34 German Version) (24) is a 34-item questionnaire and comprises of subgroups for psychological needs (10 items), health system and informational needs (11 items), physical and daily life needs (5 items), patient care and support needs (5 items) and sexuality and other problems (3 items). The items are measured on a five-point Likert scale separated into two classes: no need (scale 1-2) and some need (scale 3-5). The overall score used for evaluation is the sum of all items of the questionnaire. High SCNS-SF-34-G scores indicate the need for more support for patients. The version used is a modified version of the SCNS-SF-34-G including only 25 questions, thus the score ranges from 25 to 125.

Analysis of participants was done in accordance with the ICH E9 guidelines for data analysis considerations (25). All randomized patients were considered as part of the intention-to-treat population. Participants of the assessment group were considered as per-protocol if baseline intervention took place. For the screening group we required at least 2 additional interventions between baseline and 6 months visit. Moreover, we require for the per-protocol population that the final visit took place 6 months after their inclusion ± 2 month. Sensitivity analyses revealed that the length of the timeframe around 6 months allowed for protocol adherence had no relevant impact on our results.

We performed a multicenter, controlled, randomized, non-blinded, longitudinal study and evaluated two different strategies for capturing physical symptoms and psychosocial needs.

For this purpose, the screening arm screened for symptoms and distress by using IPOS (15) and NCCN distress thermometer (16) repeatedly in three to six week intervals (adjusted to usual tumor therapy application cycles and oncology presentations). We did not influence whether or how palliative care was adjusted according to the screening results. IPOS was chosen as a questionnaire in the screening arm, since it is a validated and widely used tool in European countries to evaluate patients’ well-being and monitor their need for care (26–28). The NCCN was selected in adherence to the preceding APM study (13).

In the assessment arm, an initial single comprehensive recording of several dimensions of quality of life using IPOS (15), NCCN Distress Thermometer (16), plus FACT-G (general and organ-specific) (17), PHQ-4 (18), and SCNS-SF-34-G (19) has been performed. The questionnaires were again selected in adherence to the preceding APM study (13).

The study concept was designed to detect symptoms or distress, so that support measures could be initiated promptly if required. The test results were supposed to be made apparent in the institutional clinical patient´s charts, and the resulting interventions were left to the discretion of the treating team.

After 6 months, a QoL assessment was performed in both intervention arms, again using IPOS, NCCN Distress Thermometer, plus FACT-G, PHQ-4, and SCNS-SF-34-G to compare QoL. We chose this observation period as a trade-off between observing long-term effects of the intervention and avoiding excessive dropout rates, based on the data from the previous APM study (13) where data was sufficiently available after 6 months but not after 12 months, due to dropout.

In addition, study centers were asked for further data from the hospital documentation system about hospital days, emergency admissions, inclusion of specialized palliative care and other supportive services up to at most six months after start of participation.

General patient data and case report form items (CRF) were recorded in an electronic format (secuTrial).

A statistical analysis plan was written, registered and signed by the principal investigators and the responsible statistical team before analysis.

Descriptive statistics are reported as numbers and proportions or median with corresponding range as appropriate. If not stated otherwise, tests were performed two-sided on a significance level of 5%. Parameter estimates are provided with corresponding 95% confidence intervals (95%-CI).

Primary endpoint was the FACT-G score at time t = 6 months. Differences in the relative intervention effect between study arms were tested non-parametrically using a two sample t-test for the nonparametric Behrens-Fisher problem (29). In addition, a multiple linear regression model for the FACT-G was fitted with the factor study arm (screening vs. assessment) and with additional influencing factors (tumor entity, study sites, and important prognostic baseline factors such as sex, age, personal living status and lost-to-follow-up-state). Primary and survival analysis were performed on the intention-to-treat population, and secondary and sensitivity analyses were done on the per-protocol population. Secondary endpoints at time t = 6 months (NCCN Distress Thermometer, IPOS, PHQ4, SCNS-SF-34-G) are evaluated analogously to the primary score; multiple linear regression models were fitted accordingly.

Survival rates within 6 months were estimated using Kaplan-Meier curves with additional 95% confidence bands. Comparison of the two intervention groups was performed using a log-rank test. Number of hospitalizations and mean length of stay were analyzed using negative binomial regression and zero-inflated Poisson-regression, respectively. Additionally, we fitted a Cox proportional hazards model to investigate additional factors.

For the assessment arm, the change of the QoL scores between initial assessment and 6 months visit was analyzed using multiple linear regression regarding the stratification factors and taking into account further covariates such as age and sex. For the screening arm, physical symptoms and psychosocial strains (measured via IPOS) were analyzed descriptively over time.

For non-parametric testing we employed a composite testing strategy and imputed missing values with the worst possible value (e.g. for FACT-G we imputed the value zero), thus associating drop-out with the least possible value for QoL. This was then interpreted as the evaluation of the full analysis set according to a worst-case approach for handling intercurrent events  (25). Additionally, we performed a complete-cases-analysis as sensitivity analysis to assess possible impairments or biases of our study results resulting from dropouts.

Differences in QoL regarding IPOS and NCCN between study arms at baseline and after 6 months, were assessed using non-parametric testing. Pre-post comparison of QoL regarding IPOS and NCCN were analysed stratified by group using paired t-testing.

For all additional analyses, missing items were imputed as proposed in the corresponding scoring guidelines. In the case of missing total scores, we imputed ten times using predictive mean matching  (30) based on study center, sex, age, entity, time of diagnosis and study arm. All data were analyzed using R version 4.2.1 (31) with additional packages mice, nparcomp, survminer for multiple imputation, nonparametric testing and survival analysis, respectively.

The study protocol was approved by the ethics committees of all 24 study sites (PI study site no. 23/2/19) and followed the Declaration of Helsinki Ethical Principles for Medical Research. SCREBEL was registered in the German Registry for Clinical Studies (DRKS No. 00017774). Patients were included after written information, clarifications of the study and written consent.

The study was sponsored by the Innovation Funds of the German Federal Joint Committee.

The primary analysis revealed no significant differences in the intervention effects on QoL (measured via FACT-G) between the two groups at t = 6 months, neither using the worst-case approach nor in a complete-case-analysis where the relative effects in intervention were estimated to be 0.504 (95%-CI: 0.458 to 0.549) and 0.509 (95%-CI: 0.431 to 0.587), respectively. Within the linear regression model, none of the included variables turned out to have a significant non-zero influence on QoL. (see Table 2 below)

With respect to the patient’s lost-to-follow-up-state we did observe only not-significantly smaller values in QoL. Moreover, we observed slightly smaller values for patients in the screening group whereas male patients seemed to have slightly higher QoL than females. Regarding the influence of age, QoL seemed to be insensitive.

In accordance with our primary analysis, non-parametric testing for differences in the intervention effect on QoL measured by the secondary scores revealed no significant results likewise, both for the worst case-approach and the corresponding sensitivity analysis. Moreover, in the linear regression models the included variables did not turn out to have significant non-zero influences on QoL even though males seemed to have a somewhat smaller stress level of -0.25 points (95%-CI: -0.47 to -0.04) measured by PHQ4 than females, p = 0.023. The tendencies of the influences of the included factors direct in the same direction as outlined for FACT-G above. The estimated differences between assessment and screening group using primary and secondary outcome questionnaires are summarized in Table 2 .

The course of estimated Kaplan-Meier survival curves up to six months is shown in Figure 2 . A log-rank test for differences in survival times revealed no significant difference in the survival distribution between the two study arms, p = 0.309.

Within regression modelling, the probability of hospital admission within 6 months for the assessment and the screening group were estimated as 51% (95%-CI: 45% to 58%) and 61% (95%-CI: 55% to 67%), respectively, and turned out to differ significantly, p = 0.033. Given a hospital admission, the length of stay in assessment and screening group was estimated as 7.2 (95%-CI: 6.9 to 7.6) and 8.0 (95%-CI: 7.6 to 8.3) days, respectively, being significantly different (p = 0.003). The number of hospitalizations within 6 months was estimated to be 0.38 (95%-CI: 0.30 to 0.48) in the assessment group and 0.51 (95%-CI: 0.41 to 0.64) in the screening group. The difference was not significant (p = 0.07).

In a linear regression model for FACT-G within the assessment group, none of the explaining factors age, sex and visit time (baseline vs. six months) turned out to have a significant influence. The difference in FACT-G after – before intervention was 2.2 (95%-CI: -3.0 to 7.3) which turned out to be no significant increase of QoL over time, p = 0.406.

Multiple linear regression of the change of IPOS with respect to baseline depending on inclusion time in the study, age and sex yielded a significant decrease over time (p < 0.001), indicating a relief of burden. Per 30 days within study inclusion, IPOS decreased by -0.54 (95%-CI:-0.84 to -0.24) points, cf. Figure 3A . Within the same period of time, psychological strains could be reduced by -0.16 (95%-CI: -0.28 to -0.05) and physical symptoms by -0.20 (-0.39 to -0.02), see Figures 3B, C , respectively.

Additionally, we analyzed the change of QoL regarding IPOS and NCCN DT over time within and between groups. Neither at baseline nor after 6 months, IPOS, its subscores for psychological strains and physical symptoms or NCCN DT differed significantly between assessment and screening group. In the assessment group, a significant improvement of QoL could only be achieved with respect to the psychological strains subscore. In contrast, the screening group exhibited a significantly better quality of life after 6 months regarding all considered scores (see Table 3 ). The distribution of QoL changes within study arms are visualized in Figure 4 .