alex trindade assoc. prof. ttu mathematics & statistics dept. ([email protected])
TRANSCRIPT
Alex TrindadeAssoc. Prof.
TTU Mathematics & Statistics Dept.([email protected])
I’m a Statistician… Your toolbox…
Problem 1. A Model for Predicting Outcomes in
Longitudinal Data (Naranjo, Trindade & Casella, Journal American Statistical Association, 2013)
• Advantages of a State-Space Approach– Flexible, handles trends over time;– Can have multivariate outcomes, covariates, and
missing data (in both outcomes & covariates);– Ease of forecasting.
State-Space Model
Outcomes @ time t = FUNCTION( X, Y)• Y: outcomes at earlier times, • X: covariates at current and earlier times.
• Lagorio et al (2006) data: 8 patients suffering from Chronic Obstructive Pulmonary Disease (COPD).
• Response: 2-vector of lung function (FVC, FEV1).
• Exogenous covariates: nitrogen dioxide and fine particulate matter.
• Time period: 32 consecutive days in winter 1999, Rome (Italy).
• Missing: 60% in response; 10% in covariates.• Main focus: prediction.
Data Analysis
Individual Forecasts
Problem 2. Smoothing Reconstructed Non-
Parametric Survival Curves (Paige & Trindade, in progress…)
• Advantages of a Saddlepoint-Based Approach– Starts from classical Kaplan-Meier weights;– Does not need user-specified tuning parameters;– Accurate reproduction of “true” curve.
• Classic dataset (c.1980): survival times (days) of 184 patients who underwent heart transplantation.
• Events: 113 died.o First died at day 0.5;o Last died at day 2878.
• Censored: 71 still alive at end of study (day 3695).
Stanford Heart Transplant Data
Your data?