Retrospective Forecasting and Long-Term Projection of Alzheimer’s Disease Spreading

Technical Summary: Retrospective Forecasting and Long-Term Projection of Alzheimer's Disease Spreading

Problem Statement
Forecasting the spread of non-communicable diseases (NCDs), specifically Alzheimer's disease and dementia, presents significant methodological challenges due to the multifactorial nature of their etiology. Unlike infectious diseases, NCDs are driven by complex, dynamic interactions of genetic, behavioral, socio-economic, and environmental factors. Traditional forecasting approaches often rely on quantifying these specific risk factors or assume process stationarity, both of which struggle with the inherent uncertainty and non-stationarity of long-term disease dynamics. Furthermore, existing models often focus on volatile annual indicators, which can obscure stable long-term trends. The authors posit that accurate forecasting is critical for strategic healthcare planning, resource allocation, and justifying research funding, yet current methods often fail to provide reliable decadal projections under conditions of uncertainty.

Methodology
The paper proposes a novel methodology based on an Integral Inflow-Outflow Model (IIOM) with stochastic parameters. This approach shifts the analytical focus from modeling individual risk factors or volatile annual case counts to modeling the dynamics of the cumulative (integral) volumes of inflow (new cases) and outflow (mortality).

Key components of the methodology include:

1. Model Framework: The disease population is treated as a "reservoir" defined by a balance equation: $I(t) = X_{inf}(t, t_0) - X_{of}(t, t_0)$, where $I(t)$ is the number of active cases, $X_{inf}$ is the cumulative inflow, and $X_{of}$ is the cumulative outflow.
2. Stochastic Rates: The model utilizes rates of change for these cumulative volumes ($r_{inf}$ and $r_{of}$) rather than raw counts.
3. Identification of Trends: A core finding of the study is that while annual rates may be volatile, the rates of the integral inflows and outflows exhibit a monotonic decrease over time. This property allows for the construction of linear trends.
4. Forecasting Mechanism:
   • The authors calculate the average annual rate of decrease in the percentage increases of inflow and outflow ($\alpha_{inf}$ and $\alpha_{of}$) over a historical window (e.g., 10 years).
   • These rates are approximated using power functions via the least-squares method.
   • Linear trends for future stochastic rates ($\tilde{r}_{inf}$ and $\tilde{r}_{of}$) are extrapolated using the formula $\tilde{r}(\tau) = r(t) - k(t)(t - \tau)$, where $k(t)$ represents the trend coefficient derived from historical data.
   • These extrapolated rates are substituted back into the discrete IIOM equations to generate 10-year and long-term projections.
5. Validation: The model's accuracy is evaluated using the Mean Absolute Percentage Error (MAPE) against retrospective data from the Institute for Health Metrics and Evaluation (IHME) for the period 1990–2021.

Key Contributions
The authors identify three primary novelties in their approach:

1. Shift in Modeling Focus: Moving from volatile annual figures to the stochastic dynamics of accumulated case volumes to reveal stable long-term trends.
2. Operationalization of Monotonicity: Identifying and utilizing the property of monotonic decrease in the rates of integral inflows and outflows as a predictable linear trend for forecasting.
3. Practical Technique: Developing a specific forecasting technique based on extrapolating these linear trends, which has been tested on fundamentally different non-stationary processes (including COVID-19 and demographic changes) to confirm robustness.

Results
The study presents two main sets of results:

• Retrospective 10-Year Forecasts: Sample forecasts were generated in 2005, 2010, and 2015 for the World and various regions. The model demonstrated high accuracy. The abstract notes MAPE values for 10-year forecasting windows in the interval [0.23%; 2.73%], while also noting specific MAPE values in the interval [0.08%; 0.75%]. Specific regional MAPE values were generally low, with the World average at 0.43% (2005 forecast), 0.2% (2010 forecast), and 0.09% (2015 forecast). The conclusion further cites a MAPE range from 0.43% to 1.79% for the retrospective forecasts.
• Long-Term Projection to 2050: Using the model-based approach, the authors project a steady increase in the absolute number of Alzheimer's cases globally. The number of cases is projected to rise from 141.2 million in 2022 to 489.5 million by 2050.
  • The proportion of cases within the elderly population (65+) remains relatively stable (0.84–0.87), suggesting that demographic aging is the primary driver rather than changes in age-specific risk.
  • Regional disparities are significant; Western Europe is projected to face the highest burden (65 million cases), followed by High-income North America (42.3 million) and Southeast Asia (16.3 million).
  • In contrast, Tropical Latin America and Eastern Sub-Saharan Africa show moderate proportions (0.72–0.76) of cases relative to their elderly populations, reflecting different demographic structures.
  • The model confirms a monotonic deceleration in growth rates, with annual growth declining from approximately 5% in the early 2020s to 2.5% by the end of the projection period.

Significance and Claims
The paper claims that the Integral Stochastic Inflow-Outflow Model offers a practical and reliable alternative to more complex, data-intensive models (such as Bayesian meta-regression or CODEm) that require precise assessment of numerous difficult-to-quantify risk factors. By acknowledging the non-stationary nature of NCD spread and focusing on the stable trends of cumulative rates, the methodology provides a robust tool for decision-making under uncertainty.

The authors conclude that this approach establishes a solid foundation for:

• Strategic healthcare resource planning.
• Justifying investments in prevention and research.
• Monitoring epidemiological trends.

The study positions the IIOM as a promising tool for long-term public health forecasting, facilitating evidence-based decision-making in the face of the growing challenges posed by non-communicable diseases, without claiming to solve the underlying biological mechanisms of the disease.