Using logic regression to characterize extreme heat exposures and their health associations: a time-series study of emergency department visits in Atlanta

Table 2 Summarized simulation study results of the performance of logic regression

Disease	True Relative Risk	Exposure Scenario	Sensitivity^a	Specificity^a	Relative bias		Relative root mean square error (RRMSE)
Disease	True Relative Risk	Exposure Scenario	Sensitivity^a	Specificity^a	Logic regression	Known H(X_t)	Logic regression	Known H(X_t)
All Circulatory System Diseases	1.01	E1	50%	99%	−0.56%	−0.01%	1.09%	0.55%
		E2	35%	98%	−0.84%	− 0.04%	1.59%	0.69%
		E3	23%	98%	−1.13%	−0.28%	2.02%	0.85%
	1.05	E1	100%	100%	−0.11%	− 0.11%	0.69%	0.69%
		E2	99%	100%	−0.03%	0.03%	0.99%	0.71%
		E3	98%	100%	−0.09%	−0.02%	1.23%	1.00%
All Renal Diseases	1.01	E1	34%	99%	−0.40%	0.36%	1.39%	1.10%
		E2	36%	98%	−0.27%	0.10%	2.03%	1.26%
		E3	26%	98%	−0.92%	−0.22%	3.03%	1.61%
	1.05	E1	93%	100%	−0.14%	0.12%	1.75%	1.21%
		E2	82%	100%	−1.00%	0.18%	3.29%	1.16%
		E3	73%	99%	−1.37%	−0.16%	3.31%	1.57%
Heat-Related Illnesses	1.01	E1	19%	99%	−0.57%	0.35%	6.58%	5.61%
		E2	29%	98%	0.30%	1.05%	12.02%	6.93%
		E3	17%	98%	0.09%	0.91%	16.21%	9.03%
	1.05	E1	33%	98%	−2.53%	−0.23%	8.43%	6.44%
		E2	30%	98%	−0.83%	0.45%	11.66%	6.40%
		E3	23%	98%	−3.07%	−0.92%	16.42%	9.10%

^aSensitivity is defined as the proportions of days assigned as exposed using the exposure metric estimated from logic regression among days assigned as exposed using the true exposure metric (E1, E2, or E3) for simulating health data. Specificity is defined similarly for days assigned as unexposed. For each simulation scenario, the sensitivities and specificities reported are averaged across 100 simulations

ISSN: 1471-2288