Data source
The GBD 2021 employed the most up-to-date epidemiological data, complemented by refined and standardized methodologies, to systematically and comprehensively quantify health losses across 369 diseases and injuries, as well as 87 risk factors, stratified by age, sex, and geographical location, encompassing 204 countries and territories. The GBD team is committed to annual updates to ensure the accuracy and relevance of their estimates [12]. The intricacies of the methodologies applied within GBD 2021 have been thoroughly documented in prior publications [13].
To address data gaps and ensure smoothness across age, time, and location, the collected data underwent modeling via spatiotemporal Gaussian process regression. This approach facilitated interpolation in regions with incomplete datasets. Furthermore, to correct for biases stemming from diverse case definitions and study methodologies across regions, a meta-regression framework incorporating Bayesian priors, regularization, and trimming techniques was employed.
From GBD 2021, we extracted estimates and their corresponding 95% uncertainty intervals (UIs) for incidence, deaths, prevalence, and DALYs attributed to MND. All rates reported herein are standardized to per 10,000 population. Additionally, the sociodemographic index (SDI), a composite indicator reflecting income, education, and fertility levels, serving as a proxy for sociodemographic development, was utilized to categorize the 204 countries and territories into five distinct groups: high, high-middle, middle, low-middle, and low, as defined by their SDI values [14].
Descriptive analysis
To gain a holistic understanding of the burden of MND, we conducted descriptive analyses at the global, regional, and national levels. Specifically, we visually presented the global trends in the number of cases, crude rate, and age-standardized rate (ASR) for incidence, deaths, prevalence, and DALYs related to MND, disaggregated by sex (both sexes, males, and females) and spanning the period from 1990 to 2021. Furthermore, we compared the number of cases and ASR for the aforementioned indicators across global, regional (comprising 54 GBD geographic regions), and national (encompassing 204 countries and territories) levels, as well as within the five SDI groups.
Trend analysis
In the Trend Analysis section, we initially employed the Estimated Annual Percentage Change (EAPC) to quantify the overarching trend in the burden of MND. Given the importance of standardization when comparing diverse groups with varying age structures or a single group experiencing temporal changes in its age profile, the EAPC-measured trend of the ASR emerges as a more robust metric for monitoring shifts in disease patterns [15]. To derive this metric, we constructed a linear regression model where the natural logarithm of the ASR (ln(ASR)) served as the dependent variable (y), and the calendar year acted as the independent variable (x). Subsequently, the EAPC was calculated using the formula (exp(β)-1) * 100%, with its 95% confidence interval (CI) also being extracted from the model [16]. In interpreting the EAPC estimates, if both the EAPC value and the lower bound of its 95% CI are greater than 0, the ASR is deemed to be in an increasing trend. Conversely, if both the EAPC value and the upper bound of its 95% CI are less than 0, the ASR is considered to be in a decreasing trend. In all other cases, the ASR is classified as stable. This approach ensures a rigorous and standardized methodology for assessing temporal trends in the ASR of MND.
Furthermore, we used age-period-cohort (APC) model to explore the underlying trends in DALYs stratified by age, period, and birth cohort. Typically, the APC model fits a log-linear Poisson model on the Lexis diagram of observed rates and quantifies the additional effects of age, period, and birth cohort. The methodological details of APC model are described in previous literature [17]. The multicollinearity between age, period, and birth cohort inevitably leads to identification issues, making it difficult to estimate the unique effects of each age, period, and birth cohort. To address this issue, the intrinsic estimator (IE) algorithm was used to estimate the coefficients of the APC model. This study employed the IE to solve the APC model, where coefficients greater than 0 indicate increased risk, and those less than 0 indicate decreased risk. The effect coefficients were transformed into natural logarithms to calculate the relative risk (RR), enabling the observation of the effects of age, period, and cohort on MND DALYs trends. The DALYs for MND and population data of each country or region were served as data input for APC model. The data was re-coded into consecutive six 5-year periods (1990–1994, 1995–1999, … , 2015–2019), consecutive 5-year age groups (0–4, 5–9, … , 90–94, 95 plus), consecutive 5-year birth cohorts (1895–1899, 1900–1904, … , 2015–2019) to estimate the overall temporal trend in incidence, prevalence, deaths, and DALYs.
Cross-country inequality analysis
To ensure evidence-based health planning, we conducted a comprehensive cross-country inequality analysis aimed at monitoring health disparities. Specifically, we employed the Slope Index of Inequality (SII) as a key metric, which was derived from regressing the country-level prevalence of the disease across all age groups against a relative position scale tied to sociodemographic development. To account for heteroscedasticity, a robust linear regression model was applied. This method utilizes iteratively reweighted least squares, giving smaller weights to observations with larger residuals, thus minimizing the influence of outliers and ensuring more stable and reliable trend estimates [18]. This approach allowed us to quantify inequalities in MND at global level and across 21 GBD regions.
Furthermore, we calculated the Health Inequality Concentration Index by numerically integrating the area beneath the Lorenz Concentration Curve. This curve was meticulously fitted using the cumulative relative distribution of populations, ordered by their SDI, and the corresponding incidence, prevalence, deaths, and DALYs attributable to the disease [19, 20]. This methodology provided a robust assessment of the concentration of health burden across nations, enabling us to identify disparities and inform targeted interventions. A negative SII/concentration index indicates that as SDI increases, ASDR decreases, and vice versa. The greater the absolute value of the SII/concentration index, the greater the degree of inequality. Their inequality value and implications are presented in Table 1.
Decomposition analysis
To gain a profound understanding of the explanatory factors underpinning the variations in MND incidence, prevalence, deaths, and DALYs from 1990 to 2021, we performed a comprehensive decomposition analysis. This analysis dissected the contributions of population size, age structure, and epidemiological changes to the observed trends [22, 23]. By disentangling these components, we aimed to quantify the specific impact of each factor on the evolution of MND burden over time.
The decomposition methodology enabled us to estimate the number of incidence cases, prevalent cases, deaths, and DALYs attributable to each factor at every location under consideration. The calculation of these metrics for each component was carried out as follows:
({X_{ay,py,ey}} = sumnolimits_{i = 1}^{20} {left( {{a_{i,y}} * {p_y} * {e_{i,y}}} right)} )(X = incidence, prevalence, deaths and DALYs)
Where the ({X_{ay,py,ey}}) represented X based on the factors of age structure, population, and specific year (y); ({a_{i,y}}) represented the proportion of population for the age category (i) of the 20 age categories in year (y); ({p_{y}}) represented the total population in year (y) and ({e_{i,y}}) represented X rate for the age category (i) in year (y).
The contribution of each factor to the change in incidence, prevalence, deaths and DALYs from 1990 to 2021 was defined by the effect of one factor changing while the other factors were held constant.
Predictive analysis
To inform the formulation of effective public health policies and the optimal allocation of healthcare resources, we conducted a predictive analysis of the MND burden in the coming decades. For this purpose, we employed the Bayesian age-period-cohort (BAPC) model, augmented with the integrated nested Laplace approximation (INLA) technique. This advanced approach, which has been shown to outperform the conventional annual percentage change model in terms of coverage and precision, was utilized to forecast the global MND burden until 2046.
The adoption of INLA within the BAPC framework offers several advantages. By approximating marginal posterior distributions, it mitigates the mixing and convergence issues that are often encountered with the Markov Chain Monte Carlo sampling techniques traditionally applied in Bayesian methods [24]. This enhancement ensures more reliable and accurate predictions of the future MND burden, thereby supporting evidence-based decision-making in public health planning.
Frontier analysis
To assess the interplay between the burden of MND and sociodemographic development, we employed a frontier analysis approach. This methodology aimed to delineate the lowest potentially attainable ASR of incidence, prevalence, deaths, and DALYs for each country or territory, contingent upon its SDI. The frontier serves as a benchmark, indicating the minimal achievable level given a country’s or territory’s development status. The deviation from this frontier, termed the effective difference, highlights potential untapped opportunities for improvement or gains, commensurate with the country’s or territory’s position on the development spectrum.
To accommodate non-linear relationships, we conducted a data envelope analysis utilizing the free disposal hull method. This analysis generated an age-adjusted frontier by SDI, utilizing data spanning from 1990 to 2021 [25]. To account for uncertainty, we implemented a bootstrapping procedure, drawing 1,000 samples with replacement from the entire dataset encompassing all countries and territories across all years. From these bootstrapped samples, we computed the mean incidence, prevalence, deaths, and DALYs at each SDI value.
Subsequently, we employed LOESS (Locally Estimated Scatterplot Smoothing) regression with a local polynomial degree of 1 and a span of 0.2 to produce a smooth frontier line [25]. This approach ensured a robust and visually interpretable representation of the frontier, while mitigating the influence of outliers. To further refine the analysis, super-efficient countries, which may distort the frontier due to exceptional performance, were excluded from the frontier generation process.