The OSCAR evaluation is a mixed-method study consisting of three research modules. Modules 1 and 2 involve a non-randomized, controlled, multi-center intervention study which compares primary data from regularly conducted questionnaires and claims data from German statutory health insurance funds in the intervention and control groups. In addition, a qualitative content analysis is planned in module 3 and will be based on the project diaries completed by each SCN. The analysis will assess OSCAR’s effectivity and the benefit of implementing it in the German healthcare system (Fig. 1).
To compare quality of life over study period between the intervention group (patients who had met an SCN) and the control group (Module 1).
To compare the health literacy, participation, and physician-patient communication in the intervention and control groups (Module 1)
To explore the associations between quality of life and health literacy, participation, and physician-patient communication (Module 1)
To compare the incidence rate of outpatient visits in both groups (Modules 1 and 2)
To compare the healthcare costs in both groups (Module 2)
To assess the SCN’s practical experiences (Module 3)
Inclusion and exclusion criteria for participation in the evaluation survey in modules 1 and 2
Module 1: primary data, patient-reported outcomes
The recruitment was carried out in four different hospitals with oncology departments across three cities in two states in Germany. Two of the hospitals are located in Berlin, while the other two are in Leverkusen and Duisburg.
In all four study sites, the SCNs recruited BKK-insured patients for the intervention group by obtaining informed consent. The SCNs are responsible for the monthly consultations to support the patients in the intervention group. Patients who are insured with a different provider were recruited by Study Nurses (SNs) for the control group. The SNs’ main task is to conduct the evaluation questionnaires in both study groups up to four times within 1 year. The evaluation’s baseline survey will be conducted after recruitment (t0) and followed up after 90 (t1), 180 (t2), and 365 (t3) days. The patients can choose between personal, telephone, and postal interviews, depending on their health status and which method is most comfortable for them. Dividing responsibilities between SCNs and SNs is necessary to ensure the independence of the evaluation surveys. Patients in both groups were recruited at the same study sites between February 2018 and the end of February 2019.
The items for the evaluation questionnaire are listed in Table 1:
Further instruments are used to measure sociodemographic variables, especially with regard to age, sex, migration status, and social status [19,20,21]. Finally, participants in the intervention group will be asked to evaluate their contact with the SCN for the last 3 months. The items are related to specific consultation topics and whether or not the support was useful for the respective topic. Participants will also be asked about the quality and quantity of contact with the SCN.
Sample size calculation
A sample size of 100 participants in the intervention group and 150 in the control group will reach a power of 80% at a two-sided level of significance of 5% to detect a difference in the EORTC QLQ-C30 (scores range from 0 to 100) with a Cohen’s d effect size of 0.4 (mean difference of ten scores and standard deviation (SD) of 25 scores) . Given that the effect size might be smaller due to a lower mean difference or higher SD, and that the severity of cancer can cause higher dropout rates, we aim to include 150 participants in the intervention group and 200 in the control group (so a total sample size of 350 participants).
The sample size calculation based on the t-test was used in spite of the intended analysis with a baseline-adjusted repeated-measures linear mixed model (ANCOVA, three-level random intercept model to account for repeated measures in patients and clustering in centers). It can be shown  that a conservative approach for estimating sample sizes has the same power as a t-test with n subjects where p is the variance deflation factor, calculated by the correlation of baseline and follow-up measures. Assuming the worst case of p = 0 leads to the sample size based on the t-test.
Statistical analysis plan
All statistical tests are performed using Stata IC15 (StataCorp, 2017, College Station, TX, USA). The primary hypothesis will be tested at a two-sided significance level of α = 0.05. All secondary hypotheses will be tested within an exploratory framework.
Descriptive statistics and the number of participants reflected in the calculation (n) will be presented in each group. For continuous variables, mean with SD for normal distribution and median with interquartile range (IQR) for other distribution variables will be presented. For categorical data, frequencies and percentages will be displayed for each category. Graphic methods such as box plots and line graphs will be used for visualizing the data.
The comparison of EORTC QLQ-C30 (global health status as a primary outcome), physician-patient communication (scores), health competence (scores), participation (scores), and knowledge (scores) over the study period between the groups will be reported as mean and 95% CI, and performed using a linear mixed model (LMM) with three levels over all available time points. Random intercepts for the patient ID and for the study clinics are included in the model to account for the cluster structure of the data. To keep selection bias to a minimum, we will develop a propensity score for adjusting baseline factors. The propensity score will be used with the inverse probability of treatment weighting (IPTW) method because the results from this method are similar to a randomized trial [23, 24].
The number of outpatient visits will be counted and the study time will be calculated over the follow-up time for each patient. The incidence rate of outpatient visits per person-time will be presented, and Poisson regression will be used to compare the incidence rate between groups.
All outcomes will be analyzed using modified intent-to-treat populations including all subjects who receive at least one consultation from an SCN and for whom at least the first follow-up assessment at Visit 1 (at 3 months) is available.
Dropouts and missing data
Reasons for dropouts will be documented and reported. If patients are alive and missings are assumed to be missing at random (MAR), we will use multiple imputation methods based on ten imputed data sets  with multiple imputation by chained equations (MICE).
Module 2: secondary data, claims data from health insurance funds
Design and participants
Another part of the OSCAR evaluation involves analyzing claims data from the BKK health insurance funds. In contrast to the patient-reported parameters in Module 1, the claims data will make it possible to analyze objective parameters. This kind of data is a valid and relevant source, and has been used in many healthcare research projects in Germany. The data from Modules 1 and 2 will not be linked. If available, the secondary data will be analyzed for the period up to 12 months before and after enrollment.
Inclusion and exclusion criteria
This data is only available for the patients in the intervention group, who have health insurance with BKK (n = 150). A new, independent and anonymous control group will therefore be drawn from BKK-insured patients who did not take part in the OSCAR intervention (n = 600). All included patients must meet the inclusion and exclusion criteria mentioned above. Furthermore, the patients in the control group should be treated in hospitals comparable to those where the patients in the intervention group were treated. In addition, the control group will be matched with the intervention group by age, gender, and diagnosis.
The following aspects are of interest: time-to-death, all costs of treatment, and the incidence rate of hospitalization and outpatient visits.
Statistical analysis plan
Paired t-test or Wilcoxon signed rank test will be performed to compare the costs of treatment per month. The generalized estimating equation (GEE) for Poisson regression will be applied to compare the incidence rate of hospitalization and outpatient visits between intervention and matched controls. Kaplan-Meier curve and Cox regression analysis will be performed to compare time-to-death between the groups.
Module 3: qualitative analyses of SCN project diaries
Design and participants
The qualitative analyses of project diaries provide important insights into the work of the SCNs. The nurses will be asked to record their feedback concerning positive aspects of the OSCAR program, as well as opportunities for further improvements with regard to the transferability of OSCAR to the regular healthcare system. For each patient, the SCNs will record the number of additional consultations between two planned monthly consultations, and any recommendations to visit partners in their supply network.
The following aspects are of interest for the qualitative content analysis: acceptance of OSCAR among the patients and among the participating hospitals; number of additional consultations for patients; subjective feedback and reasons for early program termination by the patient; barriers to consultations.