Sample
The Hamburg City Health Study (HCHS) is a population-based observational study conducted at the University Medical Center Hamburg-Eppendorf (UKE) since 2016 (in detail: [
15]). Hamburg is the second most populous city in Germany, with more than 1.8 million inhabitants.
It comprises seven districts and 104 urban quarters, accommodating people from diverse social backgrounds. Although Hamburg is predominantly urban, it also includes rural areas. Of the city’s residents, more than 0.8 million are aged 45 years or older. Hamburg’s demographic composition differs from that of Germany in several ways. Specifically, Hamburg is characterized by a higher proportion of people of different nationalities, a higher proportion of single people, and a higher proportion of residents with higher educational qualifications and income (see [
15]).
More than 30 UKE clinics and institutes collaborate to collect and analyze various risk factors for common diseases, such as heart attack, depression, cancer, stroke, and dementia. The key aims of the HCHS were to: (1) ascertain the factors contributing to the emergence of functional health limitations and major chronic conditions, (2) examine the predictive factors influencing the survival of individuals with chronic conditions, and (3) identify the factors that bolster life in individuals who have survived major chronic illnesses [
15].
A random sample stratified by age and sex was drawn from the residents’ registration office to include participants from the general population of Hamburg, aged 45 to 74 years at the time of sampling. Individuals with insufficient language or cognitive abilities to complete the questionnaires, as well as those with physical limitations that prevented them from participating in the seven-hour examination program at the study center, were excluded. As of August 2022, more than 17,000 participants underwent examinations.
In the HCHS, participants underwent a comprehensive assessment comprising 13 established and five innovative examinations focused on the function and structure of major organ systems. Additionally, the evaluation incorporated self-reported information through questionnaires on various aspects of the participants’ lives, such as, lifestyle, environment, diet, physical activity, sexual health, work life, psychosocial factors, quality of life, digital media usage, medical and family history, and healthcare utilization [
15].
In April 2020, during the pandemic, a collaboration was established between the Free and Hanseatic City of Hamburg and an interdisciplinary consortium of the UKE. As a result of this funding project, the HCHS integrated the COVID-19 module, which includes, among other things, targeted questions specifically related to COVID-19 (such as the outcomes presented in the next section).
This was a cross-sectional, observational study. All consecutive HCHS participants between April 2020 and November 2021 received the COVID-19 questionnaire. The inclusion and exclusion criteria corresponded to those of the HCHS (see also: [
16]). In sum, n = 1,840 persons were included in the analytical sample. The translated (i.e., English version) dependent variables and translated independent variable of interest (loneliness) are shown in Supplementary File
1.
All participants gave their written informed consent to participate in the study and to the analysis of the collected data. This study was approved by the Ethics Committee of the Hamburg Medical Association (PV5113).
Dependent variables
The questions regarding the utilization of healthcare services during the COVID-19 pandemic are similar to those used in other large-scale studies [
1,
17,
18]. Participants were first asked (1) whether they had personally chosen not to visit a physician despite scheduled appointments or experiencing symptoms. If they responded affirmatively, additional information was gathered about the specific types of appointments that were postponed across various medical fields, such as, (2) general practitioners, (3) specialists, and (4) dentists. Moreover, the participants were asked if they had refrained from (5) seeking treatment in the emergency room, even in the case of a medical emergency. Owing to the small number of cases, only variables (1) to (4) served as separate outcome measures (see the statistical analysis section).
Moreover, the participants were queried about (1) any cancellation or rescheduling of planned appointments by healthcare providers. Thereafter, they were prompted to indicate the nature of these appointments, including whether these were (2) specialist consultations, (3) general practitioner visits, (4) dental appointments, (5) hospital treatments or surgeries, (6) inpatient rehabilitation programs, (7) psychotherapeutic sessions, or (8) other therapeutic treatments. However, owing to the small number of cases, only the general cancellation of planned appointments by healthcare providers (1) served as an additional outcome measure. For each outcome, individuals should refer to the period from February 2020 onward.
Covariates
Covariates were selected based on the (extended [
21]) model of healthcare use developed by Andersen [
22]. This model divides predisposing characteristics (e.g., sex or age), enabling resources (e.g., health insurance status or income), and need factors (e.g., chronic conditions). Notably, in accordance with the extended model, loneliness can be treated as a psychosocial factor (in addition to the predisposing characteristics, enabling resources, and need factors).
Regarding the predisposing characteristics, we included the following in the regression analysis: age in years, sex (male and female), and marital status (married, living together with spouse; married, living separated from spouse; single; divorced; and widowed). Marital status was dichotomized (1 = married, living together with spouse; 0 = otherwise).
Regarding the enabling resources, we included in the regression analysis: household net income (17 categories) and health insurance status (statutory health insurance; other [e.g., including private health insurance]). The 17 income categories include: 1 = Under 500 €, 2 = 500 € to under 750 €, 3 = 750 € to under 1,000 €, 4 = 1,000 € to under 1,250 €, 5 = 1,250 € to under 1,500 €, 6 = 1,500 € to under 1,750 €, 7 = 1,750 € to under 2,000 €, 8 = 2,000 € to under 2,250 €, 9 = 2,250 € to under 2,500 €, 10 = 2,500 € to under 3,000 €, 11 = 3,000 € to under 3,500 €, 12 = 3,500 € to under 4,000 €, 13 = 4,000 € to under 4,500 €, 14 = 4,500 € to under 5,000 €, 15 = 5,000 € to under 6,000 €, 16 = 6,000 € to under 8,000 €, 17 = 8,000 € or higher. Based on this, we formed empirical income tertiles.
Regarding the need factors, we included a score for chronic conditions in the regression analysis. To this end, a count score for the chronic conditions (for each chronic condition: 0 = absence of the condition; 1 = presence of the condition) was calculated. The 14 chronic conditions included: arterial hypertension, diabetes mellitus, atrial fibrillation, stroke, heart failure, dementia, bronchial asthma, chronic bronchitis/COPD, myocardial infarction, atrial fibrillation, coronary artery disease, angina pectoris, cancer, and kidney disease.
Statistical analysis
Sample characteristics were provided for the total analytical sample and stratified by medical appointments cancelled by individuals (no or yes) and appointments cancelled by healthcare providers (no or yes). Subsequently, we used several penalized maximum likelihood logistic regressions [
23,
24] (using the Stata tool: “firthlogit” [
25]). To mitigate the small-sample bias caused by the limited case numbers for certain variables, the Firth method [
23] was employed. More precisely, this approach was selected to address the potential statistical challenges arising from the small sample size. List-wise deletion was applied to address the missing values. In the sensitivity analysis, the number of individuals in the household (including the respondent) was added to the main regression model.
Outcome measures included: cancelled medical appointments in general, cancelled GP appointments, cancelled specialist appointments, and cancelled dentist appointments. Appointments cancelled by healthcare providers served as the outcome measure.
The level of statistical significance was set at p < 0.05. P-values within the range of 0.05–0.10 were regarded as marginally significant. All statistical analyses were conducted utilizing the Stata 16.1 (Stata Corp., College Station, Texas, USA).