B0992
Title: A semiparametric single index model with non-Gaussian residuals for quantifying periodontal disease
Authors: Qingyang Liu - Texas A&M University (United States) [presenting]
Dipankar Bandyopadhyay - Virginia Commonwealth University (United States)
Debdeep Pati - Texas A&M University (United States)
Abstract: Periodontal disease (PD), which contributes to eventual tooth loss, remains a significant oral health burden worldwide. The escalating costs of dental healthcare in the United States, reaching a staggering US\$124 billion in 2016, necessitate the development of innovative epidemiological tools and software for accurately quantifying the risk of PD. Most of the existing epidemiological tools and software often rely on Gaussian assumptions, leading to imprecise parameter estimates for PD responses that are highly right-skewed and thick-tailed. To address this limitation, a Single Index Model (SIM) is proposed that captures the combined effect of risk factors on an individual by employing a scalar called the single index. The index represents a linear combination of the risk factors, with the coefficients' magnitude and direction indicating the relative importance of each risk factor. The proposed SIM extends the standard linear model by allowing the mean response to follow a general non-linear function of the single index while accommodating non-Gaussian residuals. The monotonic relationship between the mean response and the single index enables the use of the index to rank patients according to their PD risk, facilitating interpretation.
