The population-based “Diet, Cancer and Health” cohort includes data from 57,053 men and women aged between 50 and 64 years at their enrollment in the cohort in 1993–199718. The databank based on this cohort includes information concerning dietary habits and lifestyle obtained from self-administered questionnaires and anthropometric measurements performed by trained staff at enrollment. For the current study, we defined baseline as 1 July 1997, when the last person was enrolled in the cohort, and we included those who were alive and free of relevant diagnoses at baseline. Thus, the age range at baseline was 50 to 69 years. Follow-up was until death, emigration, or 31 December 2018, whichever came first. Establishment of the cohort was approved by the Regional Health Research Ethics Committees for the Capital Region of Denmark ((KF) 01–345/93) and by the Danish Data Protection Agency. All methods were carried out in accordance with relevant guidelines and regulations and all participants gave written informed consent.
Assessment of relevant diagnoses
Linkage with national registries was possible for all cohort members using the Danish ID number. We defined disease-related healthy aging as an absence of all cancers (ICD-10 codes: C00-C97, except C44), type 2 diabetes (E11), ischemic heart disease (I20-I25), cerebrovascular disease (I60-I64), asthma (J45), COPD (J44), and dementia (F00-F03). The specific registries are presented in the Appendix.
Exposure assessment—health behaviors
A health score was derived based on five health behaviors: smoking status, sport activity, alcohol consumption, diet, and waist circumference. The health score ranged from 0 to 9 points, with 9 points representing the healthiest behavior, hence the presumed lowest risk. Smoking status was scored as never (2 points), former (1 point), or current (0 point).
Sport activity was assessed by questions covering average weekly leisure time spent on activities such as jogging, football, and swimming during summer and winter. Activities included in this category were presumed to be of moderate to high intensity. Sport activity for a minimum of 30 min/week during the past year was considered to reflect a regular physical activity level (1 point), whereas less was considered low activity (0 point).
Both alcohol intake and dietary habits were derived from a validated food frequency questionnaire19. Alcohol consumption was estimated from intake of beer, wine, and liquor to calculate the total amount of alcohol per day (1 unit = 12 g) and scored according to former sex-specific Danish recommendations (applicable in 2010 to 2022).
Alcohol consumption was scored as maximum 7 units/week for women and 14 units/week for men as healthy intake (2 points), 7–14 or 14–21 units/week, respectively, as moderately high intake (1 point), and > 14 or > 21 units/week, respectively, for high intake (0 point). Diet was included based on four dietary recommendations with good evidence to support a role in lifestyle-related disease: whole grains, fruit and vegetables20, red meat, and processed meat21. For each item, questions containing relevant foods were summed. Whole-grain foods included oatmeal and muesli, rye bread, and whole-grain bread with the recommendation of 75 g/day (per 10 MJ/day). All fruits and vegetables (fresh and frozen), excluding juices, were summed for a daily average with the recommendation of a minimum 600 g/day regardless of the proportion between the two. For red meat (beef, pork, and lamb), the recommendation is to limit intake to a maximum 350 g/week. For processed meat, the current recommendation is to avoid or limit intake as much as possible, which we defined to be a maximum 10 g/d. The diet score ranged from 0 to 4 points, with a score of 4 given for adherence to all four recommendations: a high intake of whole grains (≥ 75 g/d/10 MJ/d) and fruits and vegetables (≥ 600 g/d), and low intake of red (≤ 350 g/w) and processed meats (≤ 10 g/d). For the composite health score, diet was grouped by scores of 3–4 (2 points), 1–2 (1 point), and 0 (0 point). Waist circumference was grouped according to official Europid sex-specific cut-off values22 for low, moderate, and high risk of cardio-metabolic diseases. Waist circumference was scored based on ≤ 80 cm or ≤ 94 cm (2 points), 80–88 cm or 94–102 cm (1 point), and > 88 cm or > 102 cm (0 point) for women and men, respectively.
Socioeconomic covariates
Information about cohabitation status and education was obtained from Statistics Denmark. Cohabitation status was based on information from household registries and defined as living with a partner or not. Education level was grouped into short (i.e., primary school), middle (i.e., high school, primary school longer than the 7 years mandatory for Danish citizens born before 1958, vocational education, supplementary education), and long (i.e., higher education).
Data on hospitalization
We quantified the hospital load by duration of hospital stays during follow-up. We identified patient entries for which the patient was physically at the hospital from among all of the patient entries in the DNPR according to algorithms from the final report of the Expert Technician Commission on Hospitalizations as appointed by the Danish National Health Data Authority23. See the Appendix for details on the definition of duration of hospital stay.
Statistical analysis
Using an illness-death multistate model, we assessed the expected length of stay in the stages “healthy” and “with disease”. The association between the health score and length of stay was estimated using pseudo-observations as the outcome in linear regression models with generalized estimating equations with a robust variance estimator24. We assumed a linear effect of health score on the length of stay, and the reported estimates are the difference in mean length of stay per 1 point increment in health score, or the per 9-point increment corresponding to comparing the most extreme health scores. As a secondary analysis, we substituted the simple health score with the five health parameters, mutually adjusted, into one multistate model. We also extended the multistate model to multimorbidity with the stages “with one disease” and “with two or more diseases” due to few observations of multimorbidity with three and more diagnoses. To illustrate these results, we predicted the mean length of stay within 20 years from baseline for 0–9 health score points for an “average person” with fixed values of cohabitation (together) and education level (middle). We carried out sensitivity analyses of the length of stay with modified versions of the simple health score by grouping the diet score into 2–4 (2 points), 1 (1 point), and 0 (0 point), allowing for a less strict definition of a healthy diet, and by substituting waist circumference with body mass index (BMI).
Hereafter, expected length of stay is referred to as life expectancy, which should be interpreted as the residual restricted life expectancy, as the length of stay in a state was defined as the time spent in a state from baseline (hence, “residual”) until 20 years after (hence, “restricted”).
The association between the health score and the continuous variable days of hospitalization per year was analyzed by a quasi-Poisson model with a log-link. The effect of the health score was modeled as a third-degree polynomial due to nonlinearity; thus, the estimates do not have an easy interpretation and were not reported. Instead, we reported the predicted mean number of days of hospitalization per year calculated from the predicted median number of days obtained from the model due to the log-link for 0–9 health score points at fixed values of the adjustment variables. We conducted sensitivity analyses of the days of hospitalization using bed days instead of calendar days.
To formally assess the linearity assumptions regarding the effect of health score and age, we visually inspected the Pearson residuals and evaluated the significance of both log and second- and third-order terms using likelihood ratio tests.
The primary results were reported for the specific ages 55, 60, and 65 years at baseline, but with a focus on the 65-year-olds because they are most susceptible to being diagnosed with disease and are closest to completing follow-up until end of life.
All analyses were adjusted for age at baseline, education, and cohabitation status and carried out separately for men and women. The effect of age was modeled as a second-degree polynomial for the models of length of stay with disease, but was otherwise modeled as linear. We included interactions between adjustment variables and the health score whenever significant in a likelihood ratio test; thus, an interaction between age and health score is included in some models. Consequently, the health score estimates from these models could be reported only for specific age values.
All statistical tests were two-sided, and we considered P < 0.05 to be significant. The statistical analyses were performed in R, version 4.3.2.