Propensity Score Matching for Infective Endocarditis

T2DM status represents the treatment indicator introduced in the theoretical part of this study. As can be seen in Table 1, significant differences are found between people with or without T2DM for age and for almost every clinical variable analyzed. Therefore, it was deemed necessary to use a statistical method, such as PSM (PSM), to control the confounding effect of baseline characteristics. The main aim of the propensity score method is to obtain an unbiased estimate of treatment effect adjusted for the impact of given confounding factors in non-randomized and observational studies [36 (link)]. The PSM method allows selecting participants with and without T2DM with the same, or nearly the same propensity score obtained with logistic regression, so that we can match the structure of the confounding factors for both group of patients [36 (link), 37 (link)]. By using multivariable logistic regression, we could estimate a propensity score for each person with and without T2DM in our investigation [36 (link)]. The variables included in the model were year of admission, sex, age, all comorbidities analyzed (Additional file 1: Table S1), previous aortic valve disease and prosthetic valve carriers.Table 1

Incidence, clinical characteristics and in-hospital outcomes of patients hospitalized with infective endocarditis in Spain from 2001 to 2015 according to T2DM status

	T2DM	2001–2003	2004–2006	2007–2009	2010–2012	2013–2015	Total	Trend
Number of endocarditis	Yes	345	486	603	962	1040	3436	< 0.001
	No	2019	2128	2304	3165	3574	13,190	< 0.001
Incidence per 100,000 per year*	Yes	6.0	7.8	8.9	12.8	13.1	10.0	< 0.001
	No	3.9	3.8	3.8	5.1	5.5	4.5	< 0.001
Age, mean (SD)*	Yes	68.7 (9.8)	69.2 (9.9)	70.4 (9.9)	71.6 (10.2)	71.8 (10.3)	70.8 (10.2)	< 0.001
	No	66.2 (12.1)	66.0 (12.3)	67.4 (12.5)	69.0 (12.2)	69.4 (12.4)	67.9 (12.4)	< 0.001
Female sex, n (%)	Yes	130 (37.7)	193 (39.7)	233 (38.6)	348 (36.2)	336 (32.3)	1240 (36.1)	0.023
	No	667 (33.0)	704 (33.1)	786 (34.1)	1037 (32.8)	1131 (31.7)	4325 (32.8)	0.393
Charlson Comorbidity Index, mean (SD)*	Yes	1.0 (0.8)	1.0 (0.8)	1.0 (0.8)	1.1 (0.8)	1.1 (0.8)	1.1 (0.8)	< 0.001
	No	0.8 (0.8)	0.8 (0.8)	0.9 (0.8)	1.0 (0.8)	1.0 (0.8)	0.9 (0.8)	< 0.001
Prosthetic valve carriers, n (%)	Yes	27 (7.8)	41 (8.4)	57 (9.5)	106 (11.0)	102 (9.8)	333 (9.7)	0.375
	No	170 (8.4)	191 (9.0)	205 (8.9)	294 (9.3)	360 (10.1)	1220 (9.3)	0.280
Previous mitral valve disease, n (%)*	Yes	99 (28.7)	122 (25.1)	156 (25.9)	271 (28.2)	264 (25.4)	912 (26.5)	0.478
	No	540 (26.8)	616 (29.0)	681 (29.6)	1034 (32.7)	1100 (30.8)	3971 (30.1)	< 0.001
Previous aortic valve disease, n (%)*	Yes	75 (21.7)	114 (23.5)	149 (24.7)	251 (26.1)	280 (26.9)	869 (25.3)	0.280
	No	511 (25.3)	602 (28.3)	673 (29.2)	972 (30.7)	1122 (31.4)	3880 (29.4)	< 0.001
Readmissions, n (%)*	Yes	58 (16.8)	91 (18.7)	116 (19.2)	199 (20.7)	223 (21.4)	687 (20.0)	0.338
	No	300 (14.9)	314 (14.8)	354 (15.4)	526 (16.6)	597 (16.7)	2091 (15.9)	0.128
Length of hospital stay, mean (SD)	Yes	28.24 (20.9)	28.9 (21.4)	28.48 (20.4)	27.69 (21.0)	24.76 (18.9)	27.17 (20.4)	< 0.001
	No	28.33 (22.6)	27.87 (21.8)	28.25 (22.4)	27.18 (22.4)	26.66 (21.4)	27.51 (22.1)	0.018
In-hospital mortality, n (%)	Yes	87 (25.2)	101 (20.8)	135 (22.4)	209 (21.7)	182 (17.5)	714 (20.8)	0.015
	No	431 (21.4)	397 (18.7)	439 (19.1)	638 (20.2)	657 (18.4)	2562 (19.4)	0.053
Costs, mean (SD)*	Yes	9137.6 (5486)	12,380.7 (8116.2)	14,928.4 (8646.3)	13,507.4 (11,465.3)	13,744.5 (10,822.8)	13,230.4 (10,000.7)	< 0.001
	No	10,171.7 (7450.2)	12,746.6 (8463.6)	16,366.4 (11,276.8)	16,588.5 (17,974.3)	14,759.3 (12,184.5)	14,452.0 (12,845.6)	< 0.001

T2DM type 2 diabetes mellitus, SD standard deviation

*Denotes a p value < 0.05 for the difference when comparing total values between patients with and without T2DM (namely, the results in the “TOTAL” column). Trends for incidence were assessed using multivariable Poisson regression models adjusted by age and sex

For PSM we used the PSMATCH2 Stata module. The Matching method chosen was one-to-one within caliper of width equal to 0.2 of the standard deviation of the logit of the propensity score [38 , 39 (link)]. Using this method, we could match all case of T2DM with a non-diabetic patient. To assess the balance of the samples after PSM we estimated the absolute standardized difference before and after matching. As can be seen in Additional file 3: Table S2 none of the absolute standardized differences after PSM were above 10%, that represents meaningful imbalance.