Table 4.

Advantages and disadvantages of different survival type methods for modeling age of motor-onset when genetic mutation status is unknown.

Method	Advantages	Disadvantages
1 Parametric model (Chen et al. 2012) 2 Nonparametric model ( Ma and Wang, 2014b)	• Extends Langbehn et al. (2004) model to genetic mixture models for kin-cohort studies. • Ma and Wang (2014b) model assumes nonparametric forms of subject-specific effects, whereas Chen et al (2012) uses same explicit parametric form as Langbehn et al. (2004). • Consistent estimator.	• Overly simplified assumption of the gamete transmission process: assumes a family member inherits the same CAG repeat-length as his proband if the proband has the gene mutation. • Ma and Wang (2014b) model is computationally demanding.
Type II Nonparametric Maximum Likelihood Estimator (NPMLE) (Chatterjee and Wacholder, 2001)	• Directly maximizes the nonparametric likelihood using an Expectation-Maximization algorithm.	• Biased and unreliable estimates of cumulative risk of onset. • Computationally demanding.
1 Type I NPMLE (Wacholder et al., 1998) 2 Independent NPMLE (Fine et al., 2004) 3 Inverse Probability Weighting (IPW) estimator (Wang et al, 2012) 4 Imputation Estimator (Wang et al, 2012) 5 Weighted least squares estimator (Ma and Wang, 2014a) 6 Isotone regression (Qin et al., 2014)	• Consistent estimator. • Resulting estimated cumulative risk curve can  - Serve as time-dependent positive and negative predictive values of the HD gene mutation test.  - Provide a numerical summary of cumulative risk associated with a positive mutation test.  - Predict the risk of onset for a subject based on his genetic test result and demographic information.  - Predict conditional probabilities of developing HD in next s-years. • Augmented IPW estimator and Imputation estimator have least variability. • Weighted least squares estimator is easiest to compute. • Isotone regression estimator is guaranteed to satisfy the mathematical properties of a distribution function.	• Type I NPMLE has high variability and its estimates of cumulative risk of onset disagrees with clinical findings. • Independent NPMLE can violate mathematical constraints on probability risk. • IPW has high variability. • IPW, Augmented IPW are susceptible to division by zero. • IPW, Augmented IPW, Imputation and Weighted least squares estimators are not guaranteed to satisfy the mathematical properties of a distribution function and may violate the constraints of a probability where values must be between zero and one. • Imputation and isotone regression estimator are computationally demanding.