Expanding behavior pattern sensitivity analysis with model selection and survival analysis

BMC Veterinary Research

Table 2 Linear regression models for time to proportion-resistant equilibrium of the three behavior modes

Behavior Mode	Behavior Pattern Measure	Most Parsimonious Model														Full Model
		Standardized Input Parameters Coefficient (Standard Error)											Fit Statistics			Fit Statistics
		p _r		start _r	λ _in	λ _out	δ	MIC _s	H _i	V _LI	η _LI	γ _LI	AIC	BIC	Adj R²	AIC	BIC	Adj R²
Increasing	Equilibrium time	1868 (347)		− 883 (313)		− 1085 (352)			− 1526 (352)	− 884 (321)		− 1141 (292)	427	437	0.733	436*	464*	−0.706*
Decreasing	Equilibrium Time	p _r	− 1527 (514)	1233 (207)	− 3488 (396)		− 750 (175)	− 1107 (180)		− 664 (179)			2141	2172	0.561	2149	2226	0.583
		p _r ²	805 (408)
		p _r ³	552 (331)
		p _r ⁴	− 410 (156)
Peaked	Equilibrium Time	810 (175)		− 667 (174)	− 1807 (279)	660 (139)	− 781 (123)	− 1393 (154)		− 842 (145)	− 628 (146)		3812	3846	0.423	3824	3918	0.436

The example mathematical model was for the proportion of tetracycline-resistant enteric Escherichia coli in a beef steer during and after administration of oral chlortetracycline. A separate linear regression model was built for each behavior pattern measure of each behavior mode. The behavior pattern measure was the dependent variable in the linear regression models. Equilibrium was reached by 36% of increasing behavior simulations, 38% of decreasing behavior simulations and 24% of peaked behavior simulations after chlortetracycline administration ended and before the end of the simulation period. Simulations that did not reach equilibrium were excluded from these models. Coefficients and standard errors are listed for the standardized parameters that were included in each most parsimonious linear regression model. Full model refers to a linear regression model including all the parameters listed in Table 6 as independent variables. Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and adjusted R² are given for the most parsimonious and the full model. *Excludes log₁₀(β_ir), log₁₀(β_sr), γ_s and MIC_i to prevent overfitting

ISSN: 1746-6148