Projections of disability-adjusted life years for major diseases due to a change in vegetable intake in 2017–2040 in Japan

We followed the methodological approach of Nomura and Yoneoka et al. [14]. GBD data from 1990 to 2016 were used to predict the future values from 2017 to 2040 for DALYs attributable to three disease groups that have been identified to be associated with low vegetable intake. DALYs is a comprehensive measure of disease burden, comprised of years lived with disability and years of life lost due to premature death. The three disease groups were neoplasms, CVDs and DKDs from level 2 in the GBD hierarchical causal structure. The GBD hierarchy causal structure ranges from level 1 of the three basic groups (communicable; maternal and neonatal conditions, and nutritional deficiencies; non-communicable diseases; and injuries) to level 4 of the most detailed 359 diseases groups.

Following the GBD’s prediction methodology [15], we developed a three-component model of cause-specific DALYs for the three disease groups. This model included a component explained by changes in major behavioural and metabolic risk predictors including vegetable consumption, which is the main predictor of interest in this project, socio-demographic index (SDI), and an autoregressive integrated moving average (ARIMA) model that captures the unexplained components over time. Separate projections models were developed by sex and three age groups (20–49, 50–69, and ≥ 70 years,and all ages). The 0–19 years age group was not included in the model because of a lack of data on risk predictors including smoking and alcohol intake. Further detailed information on data and model formula is described below.

We used the estimated DALY rates per 100,000 population for CVDs, all cancers, and DKDs, as well as SDI in Japan for the years 1990–2016 reported in GBD 2017 [16]. SDI is a composite indicator of development status strongly correlated with health outcomes, wherein a 0 to 1 index value was determined for each of the original three covariate inputs, including total fertility rate under the age of 25, mean education for those aged 15 and older, and lag distributed income [16]. An index score of 0 represents the minimum level of each covariate input, while an index score of 1 represents the maximum level of each covariate.

We used consecutive nationwide data from NHNS to characterise sex- and age-specific average daily vegetable intake and prevalence of current smoker, current alcohol drinker, and obesity (defined according to the Japanese definition of a body mass index (BMI) of 25 kg/m² or over) following the design of a previous study [14]. The individual data from NHNS were categorised according to the GBD’s category and combined with DALYs rate by age, sex and survey year. The NHNS is a nationally representative household survey, which is conducted annually by the Japanese Ministry of Health, Labour and Welfare in order to capture the distribution of dietary habits, nutrition intake and lifestyle at a population level [17]. Residents aged ≥1 years-old in all households from Census enumeration areas are selected using a stratified single-stage cluster sample design. The NHNS consists of three parts: 1) physical examination including a blood test performed by a medical team at community centres; 2) an in-person survey of a weighted single-day dietary record of households investigated by registered dieticians who visit and check the participant’s survey compliance; 3) a self-reported lifestyle questionnaire such as smoking status and alcohol consumption. Only those aged 20 years or older were subjects to the lifestyle questionnaire. Details of survey design and procedures of NHNS are available elsewhere [11, 17]. The dietary intake survey was conducted on a single designated day by household representatives, usually by those who are responsible for food preparation. Trained interviewers (mainly registered dietitians) instructed household representatives on the measurement of food and beverages consumed by the household members and verified the survey compliance. The proportion of shared dishes, food waste, and foods that were eaten out were also recorded. The nutrient intake and food consumption were estimated using the dietary record and the corresponding food composition list of the Japanese Standard Tables of Food Composition [18] We also obtained individual-level data on the intake of green and yellow vegetables, other vegetables, vegetable juices and salted vegetables. The data on food consumption from 1995 was used because individual food consumption values before 1994 were not available.

In this study, vegetable intake was defined as the intake of green and yellow vegetables and other vegetables by referring to GBD 2017 definition as follows: average daily consumption of vegetables (fresh, frozen, cooked, canned or dried vegetables excluding legumes and salted or pickled vegetables, juices, nuts and seeds, and starched vegetables such as potatoes or corn) [7].

This study was performed in accordance with the Declaration of Helsinki. The Research Ethics Committee of the Graduate School of Medicine, The University of Tokyo approved this study (authorization number 11964) and waived the need for informed consent as this study was a secondary analysis of anonymised data that is collected routinely by the MHLW. Data of GBD 2017 are also secondary, aggregated estimates by country, sex, and age groups.

The ARIMA model, one of the most classic methods of time series analysis, was employed to forecast future DALY rates adjusting for several risk predictors. It is a moving average (MA) model combined with an autoregression (AR) model to fit the temporal dependence structure of a time series using the shift and lag of historical information.

The Box–Jenkins methodology was adopted to fit the ARIMA (p, d, q) model with orders p, d, q: where y_t is the outcome of interest, ε_t is an (white noise) error term with an intensity of σ² at time t, L is time lag operator defined as L^ky_t = y_t − k, and α_i and β_i are the coefficient parameters [19]. Before constructing the model, the stationary state of observed data in the series must be identified using Dickey-Fuller test [19]. All variables were log-transformed. If non-stationary is assumed to be plausible, the data was transformed into a stationary time series by taking a suitable difference with order d. The autocorrelation function and the partial autocorrelation function was used to identify the stationary status and the search range for orders of the model. The model estimation was carried out after an initial model has been identified; generally, model parameters are estimated by using maximum likelihood methods. Akaike’s Information Criterion (AIC) was calculated to select optimal models with the orders.

We used a two-step approach to forecast the DALY rates: the first step was to independently forecast the values of each predictor from 1995 until 2040 using Eq. (1), and then the second step was to forecast the log-scaled DALY rates by using the Eq. (2) after inserting the predicted values of the above predictors into the model as illustrated below: where y_t was the DALY rate at time t, x_tj was the value of j th predictor at time t and γ_j was a coefficient parameter for j th predictor. All analyses were conducted by STATA version 16 using the ARIMA procedure. The parameters in Eq. (2) were separately estimated by age and sex groups.

We assumed three scenarios of better, moderate and worse cases to evaluate the impact of change in vegetable consumption on the DALY rates for three diseases (CVDs, cancer, and DKDs) for 2017–2040 using the smoothed data between 1995 and 2016. A reference forecast was set as the current trend was maintained, namely the projected vegetable consumption during 1995–2040 derived from the ARIMA model in Eq. (1). The better case scenario considered achieving the target daily vegetable consumption (350 g/ day) in 2023 as per Health Japan 21 (the second term), which the guideline provides as the recommendation and goals of lifestyle to improve population health defined by the Japanese government. This scenario assumed a constant monotonic increasing function from 2017 to 2023 and a constant level of 350 g/day after 2023. The moderate case scenario assumed that the target vegetable consumption would be achieved in 2040 rather than in 2023 with the monotonic increase function. As such, vegetable consumption as of 2040 was the same values in the better and moderate case scenarios. The worse case scenario assumed a constant monotonic decrease in vegetable consumption from 2017 to 2040 by the level of 2004 when the lowest consumption was recorded in decades because of higher vegetable prices. By inserting values derived from assumed alternative scenarios into the Eq. (2) as a predictor, we predicted the final value of DALY rates until 2040 for each scenario. The DALY rates in the better and the moderate case scenarios mathematically converged to the almost same values in 2040 since both scenarios would meet the same level of vegetable consumption by 2040.