TL;DR
This study establishes the first comprehensive Amyotrophic Lateral Sclerosis (ALS) registry in Singapore, the SingALS, and uses it to develop predictive models for patient survival. By comparing this cohort to international registries and employing machine learning, the researchers identified key demographic and laboratory factors that influence ALS progression and mortality, offering a more nuanced understanding of the disease in an Asian population.
What Problem Does This Paper Address?
Amyotrophic Lateral Sclerosis (ALS) is a devastating neurodegenerative disease with varied progression and outcomes. While international ALS registries have been instrumental in identifying prognostic markers, most research has focused on Caucasian populations. This leaves a significant gap in understanding ALS in diverse ethnic groups, particularly in Asia, where genetic backgrounds and environmental factors may influence disease trajectory. Furthermore, existing predictive models may not be directly applicable to these populations. This paper addresses this by creating a dedicated Singaporean ALS registry (SingALS) to build and validate predictive models tailored to an Asian cohort, aiming to improve prognostic accuracy and clinical management.
Methods
The researchers established the SingALS registry by retrospectively collecting data from two major Singaporean hospitals (Tan Tock Seng Hospital and Singapore General Hospital). This combined dataset included demographic, clinical, and laboratory information from 143 ALS patients. They then performed comparative analyses with international ALS registries to identify geographical and ethnic differences in ALS presentation and progression.
For predictive modeling, the study employed the eXtreme Gradient Boosting (XGBoost) algorithm, a powerful machine learning technique known for its accuracy in handling complex datasets. This model was used to predict mortality at various time points (6, 12, 24, and 36 months) using features from the TTSH ALS database, which contained more detailed longitudinal data. The model's performance was evaluated using standard metrics like AUC and AUPRC.
Additionally, Kaplan-Meier (KM) survival analysis was used to visualize survival trajectories across different demographic groups (gender, progression speed), and Cox regression analysis was employed to assess the impact of individual variables on survival while accounting for confounders. A novel temporal dynamics analysis was also conducted to examine how specific features changed over time in patients who died within 24 months versus those who survived longer.
Datasets
- SingALS Registry: A newly established comprehensive registry comprising retrospective data from 143 ALS patients from Tan Tock Seng Hospital (TTSH) and Singapore General Hospital (SGH) in Singapore. This dataset includes demographic, clinical, and laboratory data.
- International ALS Registries: Data from several international ALS registries were used for comparative analysis, including:
- German Swabia registry
- Emilia Romagna region ALS data (Italy)
- ENCALS (European Network for the Cure of ALS)
- University Malaya database (Malaysia)
- National Institute of Mental Health and Neurosciences database (India)
- Peking University database (China)
- Taiwan database
Key Insights and Ideas
- Unique Asian ALS Profile: The Singaporean ALS cohort exhibits distinct characteristics compared to Caucasian-dominant registries, including longer average survival, younger age of onset, and a lower prevalence of bulbar onset. This highlights the importance of ethnicity and geographical factors in ALS presentation.
- Male Patients Face Worse Prognosis: Consistent with some previous studies, male patients in the TTSH cohort showed poorer survival outcomes compared to females, even after adjusting for other factors.
- Early Clinical Milestones Predict Mortality: Patients who died or required tracheostomy within 24 months of symptom onset experienced significantly earlier needs for being bedbound, PEG insertion, and non-invasive ventilation (NIV). This suggests these milestones are critical early indicators of rapid disease progression.
- Inflammatory Markers as Key Predictors: Elevated levels of alkaline phosphatase (ALP), white blood cell (WBC) count, and absolute neutrophil counts emerged as significant predictors of worse mortality. This points to the role of inflammation in ALS pathogenesis and progression.
- ALSFRS-R Slope Remains Crucial: A steeper decline in the ALS Functional Rating Scale-Revised (ALSFRS-R) score, indicating faster functional loss, was consistently identified as a key predictor of poorer survival, reinforcing its established prognostic value.
- Machine Learning Uncovers Subtle Patterns: The XGBoost model was effective in identifying predictive features, including both clinical and laboratory markers, and demonstrated better performance for short-term mortality prediction. This suggests machine learning can uncover complex relationships that might not reach statistical significance in traditional regression models with smaller cohorts.
Key Results
- Singaporean ALS patients had longer average survival (50.51 months) compared to the German Swabia registry (31.0 months).
- The hazard ratio (HR) for males compared to females was 3.12 (p=0.008), indicating a significantly higher risk of mortality for males.
- Patients who died within 24 months required bedbound status (p<0.004), PEG insertion (p=0.004), and NIV (p<0.001) significantly earlier.
- Machine learning and statistical analyses identified steeper ALSFRS-R slope, higher ALP, WBC, absolute neutrophil counts, and creatinine levels as associated with worse mortality.
- The study found that male patients progressed to major clinical milestones faster than females, though this difference was not statistically significant in all analyses.
Limitations and Future Work
The authors acknowledge several limitations:
- Sample Size: While the SingALS registry represents a significant proportion of ALS cases in Singapore, the overall sample size (143 patients) is still relatively small, which may have impacted the statistical power for some analyses, particularly for less common variables or subgroups.
- International Comparisons: Direct comparisons with international registries are challenging due to variations in healthcare costs, availability of specialized services, and cultural differences in healthcare-seeking behavior.
- Data Availability: The SGH cohort's data had limitations in longitudinal follow-up and laboratory parameters, necessitating the primary analysis on the TTSH cohort for certain aspects.
Future work suggested includes:
- Expanding the Registry: Continuing to build the SingALS registry to include more patients and prospective data to enhance the predictive capabilities of the machine learning models.
- International Validation: Validating the developed models in larger, international, population-based studies, especially across diverse Asian populations, to confirm findings and improve generalizability.
- Investigating Mechanisms: Further research into the underlying mechanisms of ALS pathophysiology, particularly the role of inflammation and immune responses, to identify potential therapeutic targets.
- Exploring Financial and Cultural Impacts: Investigating how financial burdens and cultural factors influence ALS treatment access and outcomes in different regions.
Summary updated