Abstract
Though they may offer valuable patient and disease information that is impossible to study in a randomized trial, clinical disease registries also require special care and attention in causal inference. Registry data may be incomplete, inconsistent, and subject to confounding. In this paper we aim to address several analytical issues in estimating treatment effects that plague clinical registries such as the Emory amyotrophic lateral sclerosis (ALS) Clinic Registry. When attempting to assess the effect of a surgical insertion of a percutaneous endoscopic gastrostomy (PEG) tube on body mass index (BMI) using the data from the ALS Clinic Registry, one must combat issues of confounding, censoring by death, and missing outcome data that have not been addressed in previous studies of PEG. We propose a causal inference framework for estimating the survivor average causal effect (SACE) of PEG, which incorporates a model for generalized propensity scores to correct for confounding by pre-treatment variables, a model for principal stratification to account for censoring by death, and a model for the missing data mechanism. Applying the proposed framework to the ALS Clinic Registry Data, our analysis shows that PEG has a positive SACE on BMI at month 18 post-baseline; our results likely offer more definitive answers regarding the effect of PEG than previous studies of PEG.
Introduction
Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease that often leads to dysphagia, malnutrition, and rapid weight loss. Percutaneous Endoscopic Gastrostomy (PEG) tubes are frequently recommended to maintain nutritional status, yet their long-term efficacy on Body Mass Index (BMI) has remained scientifically inconclusive. The gold standard for clinical evidence, the Randomized Controlled Trial (RCT), is often unethical in this context, as withholding nutritional support from starving patients is unacceptable.
Consequently, researchers must rely on observational registry data. However, these datasets present three significant hurdles: confounding (where sicker patients are more likely to receive PEG), censoring by death (the inability to measure BMI in deceased patients), and general missingness in registry records. This paper addresses these challenges by developing a Bayesian causal inference framework to estimate the true impact of PEG.
Methodology
The proposed framework integrates three advanced statistical components to address the inherent biases in observational ALS data:
- Generalized Propensity Scores: To account for the fact that PEG can be administered at various times post-diagnosis, a Cox proportional hazards model is used to calculate propensity scores. This balances baseline characteristics—such as age, lung function, and functional status—between treated and untreated groups.
- Principal Stratification: This technique addresses "truncation by death" by categorizing patients into latent strata based on their potential survival under both treatment conditions. The primary interest is the Survivor Average Causal Effect (SACE), which measures the treatment effect specifically within the "Always Survivors" group (those who would survive to the 18-month mark regardless of whether they received PEG).
- Missing Data Mechanism: The framework incorporates a model for missingness that allows the probability of missing data to depend on the unobserved principal strata, providing a more robust fit than standard missing-at-random assumptions.
The researchers utilized a Data Augmentation (DA) algorithm and Markov Chain Monte Carlo (MCMC) sampling to iteratively impute missing strata and BMI values, allowing for a full Bayesian estimation of the treatment effect.
Data and Results
The study analyzed data from the Emory ALS Clinic Registry (1997–2014), involving 815 patients. Baseline variables included BMI, Vital Capacity (VC), and the ALS Functional Rating Scale-Revised (ALSFRS-R). The primary outcome was BMI at 18 months post-baseline.
The "Naive" Trap vs. Causal Inference
A simple, unadjusted comparison of the registry data suggested that patients receiving PEG had a BMI approximately 1 unit lower than those who did not. This "naive" result would erroneously suggest that PEG is harmful. However, this was a result of selection bias: sicker patients with lower baseline health were more likely to receive the intervention.
Key Findings
- Positive Treatment Effect: After adjusting for confounding and censoring by death, the model revealed a significant positive SACE. For the "Always Survivor" group, PEG was associated with an increase of approximately 3 BMI units at 18 months compared to the untreated state.
- Model Performance: Using the Deviance Information Criterion (DIC), the authors confirmed that a model assuming "latent ignorable missingness" (where missingness is tied to survival strata) provided the best fit for the complex registry data.
Discussion and Conclusion
This study demonstrates that when properly adjusted for survival and confounding, PEG provides a substantial nutritional benefit to ALS patients. However, some limitations remain. The registry tracks the insertion of the tube but not the actual caloric intake or adherence to the feeding regimen. Furthermore, the results rely on the assumption of "Strong Ignorability," meaning no unmeasured variables influenced both the treatment decision and the outcome.
Future research should apply this framework to other palliative interventions, such as Non-Invasive Positive Pressure Ventilation (NIPPV), and seek to incorporate more granular data on treatment adherence to further refine these causal estimates.
Summary updated