Survival Analysis of Concrete Highway Bridge Decks in Oregon Utilizing LASSO and Stepwise-Variable Selection

Abstract

This research studied the relationship between Oregon national bridge inventory (NBI) data as well as additional design/detailing variables and concrete highway bridge deck performance. Cox proportional hazards regression using least absolute shrinkage and selection operator (LASSO) and stepwise-variable selection were employed to determine the contribution of independent variables gathered from the NBI database. The stepwise approach produced a model with fewer variables compared with when LASSO was used. The hazard ratios (HR) of those variables retained in the stepwise approach were consistent with HR obtained from LASSO. The HR were also found to be intuitive. For example, cast-in-place bridge decks on steel bridges without a wearing surface or deck protection have an increased risk of deterioration while bridge decks in dry climate zones far from the ocean on prestressed concrete bridges have a decreased risk of deterioration. A separate refined dataset of 400 bridge decks was assembled by manually going through bridge drawings to gather information on relevant design/detailing variables. Bridge decks were carefully selected using a stratified sampling procedure based on design period and climate zone. Kaplan-Meier survival curves were constructed to study the impact of these design/detailing variables on concrete highway bridge deck performance. For example, bridge decks with concrete cover less than 1 in. (25 mm) have a lower survival probability than bridge decks with more cover and bridge decks that are slender have a higher survival probability than other bridge decks.