Research setting and data
Like the health care systems in countries such as the United Kingdom, Ireland and the Netherlands, the German health care system is also exploring ways to enable health professionals to refer patients to local, non-medical health services provided by community and voluntary sector organisations. Our analyses use data from the first such service in Germany to which GPs and specialists can refer patients by means of a social prescription. The community health advice and navigation service opened on 26 September 2017 and is part of a patient-oriented and cross-sectoral healthcare network aiming to improve healthcare in a district of northern Germany. The district is home to a multi-cultural population of approximately 110,000 people and has a moderate level of deprivation. The organization through which the community health interventions are provided is a privately owned regional management company (German entrepreneurial company with limited liability). It is not run as a registered charity. The management company received three years of start-up funding from the Innovation Fund of the German Federal Joint Committee (G-BA) and is now financed by several statutory health insurers. While the legal form is a for-profit company, the company has a non-profit distributing mission and it’s financing through the statutory health insurers speaks for considering it part of the voluntary and community sector, as are community service providers in other settings and countries. The service in our study can be understood as a one-stop-shop for health-related information, disease prevention, and health promotion. It offers one-to-one sessions with nurses and allied health workers, where patients can receive health advice and education (e.g., dietary change, smoking cessation, social and family issues, preparation of administrative procedures) in their mother language to better understand their individual burden of disease and opportunities for prevention. The nurses and allied health workers at the service cover eight of the most spoken languages in the neighbourhood. While this type of service is highly tailored and personalised to individual needs, the community health advice and navigation service in our study also offers group interventions. These group interventions have a more generic orientation and address specific patient groups (e.g. obese patients or patients with mental health problems). Both types of services are provided at the premises of the health advice and navigation service. In addition, similar to social prescribing in other countries such as the UK, the health care navigation and advice service involves and cooperates with existing local community services/groups/activities. Specifically, the service informs patients about the local community services and supports patients in finding and arranging appointments with the appropriate health professionals, welfare institutions, exercise groups and wellbeing groups. To determine which types of service an individual needs, nurses and allied health workers who are employed at the community health advice and navigation service undertake a comprehensive needs assessment and build-up a tailored care plan for the patient during their first consultation. For patients with a referral, the information on diagnosis or suspected disease and treatment recommendation/plan provided by the GP/specialist on the referral form is considered in this step. While there is no formal link worker as in other settings, like the UK, the nurses and allied health workers who are employed at the community health advice and navigation service partly take over the link worker functions, such as patients’ need assessment and coordinating with existing local community services/groups/activities. The service is free at point of use for all patients, regardless of whether they have a social prescription or self-refer.
The fact that patients can use the service either by obtaining a social prescription from a doctor or by self-referring allowed us to estimate directly the impact of social prescribing on patients’ adherence to the service. In this context, it should be noted that GPs in Germany do not function as gatekeepers to secondary care, the majority of which is provided in office-based practices and can be accessed by patients directly without a referral. Both GPs and consultants/specialist doctors were able to refer patients to the health care navigation and advice service examined in this study.
Data collection started on 1 January 2018. To measure service utilization, we created a cross-sectional data set by merging several sources of data maintained by the health advice and navigation service and using an ID for each of its visitors. The data set accounts for all patients who visited the service at least once from 1 January 2018 to 31 December 2019. To ensure comparability, we excluded patients younger than 18 years from our analyses (n = 110), yielding a final set of data from 1,734 patients who visited the service at least once in the observation period. The study was approved by the Institutional Review Board of the University of Hamburg.
Measures
Table 1 gives a description of the study’s variables.
Dependent variable: We used the number of times patients attended the health advice and navigation service during the three months following their initial visit (henceforth described as the number of return visits) as a proxy for patient adherence to it within the observation period. A similar understanding of patients’ adherence to such services has been used by other authors in the field of social prescribing, including Pescheny et al. [24] and Husk et al. [15]. As our observation period, we used the mean time between patients’ initial and last visit in our sample, which was 3 months. Because there is no common definition of, or empirical evidence on, the optimal length of an observation period for counting return visits, we conducted sensitivity analyses using observation periods of different lengths to test the robustness of our results (see sensitivity analyses in the results section).
Independent variables: The main variable of interest was “social prescription”, which was a binary variable and equal to one if a patient was referred to the service by a primary or secondary care professional or zero if the patient had self-referred to the service.
We controlled for variables suggested in prior research to be relevant to patient engagement and adherence, namely gender [21, 29–31], age [21, 32, 33], geographical distance to the service [15] and type of illness or reason for using the service, i.e. overweight or obesity or a psychological concern [34, 35].
Table 1
Description of study variables
Variable
|
Description
|
Number of return visits
|
Number of times a patient visited the health advice and navigation service within three months after the initial visit Count variable
|
Social prescription
|
Binary variable 0 = Patient self-refers to the service 1 = Patient is referred to the service by a doctor
|
Gender
|
Binary variable 0 = Female 1 = Male
|
Age
|
Age in years (continuous variable)
|
Distance
|
Geographical distance in km from the patient’s residence to the service (continuous variable)
|
Visit due to overweight
|
Binary variable 0 = Reason for visiting the service was not overweight or obesity 1 = Reason for visiting the service was overweight or obesity
|
Visit due to psychological concern
|
0 = Reason for visiting the service was not a psychological concern 1 = Reason for visiting the service was a psychological concern
|
Statistical analyses
Because our dependent variable (i.e., the number of return visits) is count data, we compared the fit of four models: the simple Poisson regression model, the zero-inflated Poisson model, the simple negative binomial model and the zero-inflated negative binomial model. First, we used a likelihood ratio test and assessed the over-dispersion parameter a to investigate whether a negative binomial model would be more appropriate for our analysis than a Poisson model. If a is significantly greater than zero, then the data are over-dispersed and are better estimated using the former [36]. Indeed, this was the case in our dataset (a= 0.718, p<0.01). Second, we compared model residuals and different model fit statistics to assess whether a zero-inflated negative binomial model would provide more precise estimates. For this purpose, we used a graphical comparison of model residuals, a chi² test to compare the actual distribution of the data and the distribution proposed by the models, and the Bayesian Information Criterion (BIC). The results of all tests suggested that the negative binomial model was the most appropriate for our analysis. Lastly, a chi² goodness-of-fit test showed that there was generally no evidence of misspecification in our negative binomial model. We provide further information on the model fit statistics in Appendix A. To investigate whether the social prescription effect differed depending on patient characteristics, we also tested interaction effects of the variable social prescription with the control variables.
To facilitate interpretation of our results, we transformed coefficients into incidence rate ratios (IRR). The IRR is defined as eβ and represents the average number of return visits to the health advice and navigation service with respect to a one-unit increase in the explanatory variables used in this study. In anticipation that we might have misspecified the true (but unknown) population density, we chose robust standard error procedures.
Sensitivity analyses
To test the robustness of our results, we performed two sets of sensitivity analyses.
First, we tested the robustness of our results using periods of one, two, four, five and six months.
Second, patients referred to the health advice and navigation service by a doctor might have had more complex needs, and therefore made a larger number of return visits, than patients without such a referral. To rule out this form of selection bias, we conducted a subsample analysis using data from the 197 patients for whom we had health status data and included several health-related variables, i.e. body mass index, sports activity, smoking, diabetes, hypertension and back pain. We used two-sample t-tests for continuous variables and chi²-tests for binary variables to compare the two groups: patients with a social prescription and patients who self-referred.
We performed all analyses with STATA version 15.1 (College Station, TX).