COMBINING ISOTONIC REGRESSION AND EM ALGORITHM TO PREDICT GENETIC RISK UNDER MONOTONICITY CONSTRAINT.

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Estimating disease cumulative risk from family history is challenging due to unknown genetic statuses and right-censored age-of-onset data.
• Current methods for cumulative risk estimation from family data are limited, often providing only point estimates and lacking monotonicity or non-negativity guarantees.

Purpose of the Study:

• To develop a novel, robust method for estimating cumulative disease risk across the entire time span.
• To address limitations of existing methods by incorporating probabilistic mutation status and censored data.
• To apply the new method to Parkinson's disease (PD) to analyze age-at-onset distributions in relation to PARK2 mutations.

Main Methods:

• Combined Expectation-Maximization (EM) algorithm with isotonic regression for cumulative risk estimation.
• Developed an estimator that ensures monotonicity and satisfies self-consistent estimating equations.
• Applied the method to family-based data, specifically analyzing Parkinson's disease age-at-onset in PARK2 mutation carriers and non-carriers.

Main Results:

• The novel method provides monotonic and non-negative cumulative risk estimates across the entire support.
• The estimator demonstrated high power in detecting differences in cumulative risks between populations.
• Analysis of Parkinson's disease data revealed a significant difference in age-at-onset distribution between compound heterozygous PARK2 carriers and non-carriers.

Conclusions:

• The developed Expectation-Maximization and isotonic regression method offers a significant advancement in estimating cumulative disease risk from complex family history data.
• This approach accurately models age-at-onset distributions and identifies specific genetic risk factors, as demonstrated in the Parkinson's disease study.
• The findings highlight the utility of the new method for genetic studies, particularly in understanding disease progression in relation to specific gene mutations like PARK2.