Here we found increasing incidence rates for IE in the Spanish population during the period 2001–2015, which were over twofold higher among the people with T2DM than among their non-diabetic counterparts. Over time, mean LOHS was around 27 days, and fell significantly in both people with and without T2DM, whereas costs increased significantly. IHM was around 20% both in the people with and without T2DM. Finally, in our study T2DM was not associated with a higher IHM during admission for IE.
Other authors have also identified increasing incidence rates for IE . The burden of comorbidity of an ageing population, a progressively higher number of invasive procedures, and other factors like current trends to prescribe antibiotics to prevent IE less often have been underscored by these researchers as possible reasons. However, the results from these publications are difficult to reconcile with other studies that have not detected increasing incidence rates for IE . T2DM has been proposed to induce endothelial dysfunction, which can promote bacterial adhesion and in consequence predispose patients to the onset of IE [14, 15]. Additional mechanisms that could potentially contribute to incident IE in T2DM are the impaired immune response reported in older studies among people with diabetes , or the higher prevalence of certain types of infection among the people with diabetes that are known to generate bacteremias .
We found that S. aureus and Enterococci were more frequent among T2DM patients and Streptococci among those without T2DM. The higher incidence of S. aureus infection among patients with DM and IE is consistent with previous studies [12, 23, 24]. Two probable causes suggested for this are that DM patients use healthcare services more frequently and thus have a higher chance of being exposed [6, 8, 17] and that diabetic patients have relatively higher risk for skin and mucous membrane bacterial infection compared to nondiabetic patients [12, 19, 23]. Lin et al. found a higher rate of S. aureus-related IE in patients with DM (41.7% in patients with DM vs. 27.9% in patients without DM; p = 0.021). The frequencies of other microorganism-related IEs were not different between the two groups . Kourany et al. compared microorganisms among diabetic and nondiabetic patients suffering IE and reported that S. aureus was isolated more often (30.7% vs 21.7%, p = 0.02), and from the viridans Streptococcus group less often (16.7% vs 28.2%, p = 0.001) in the diabetic group . Olmos et al.  described that enterococcus and S. bovis were more frequent in patients with DM, whereas the viridans group streptococci were more commonly isolated in those without. Chirillo et al.  analyzed 309 episodes of IE, of whom 70 had DM and Enterococcus was more commonly isolated in this last group.
Higher valve surgery rate in patients without DM before (13.9% vs 19.9%) and after matching (13.9% vs 17.3%) was found in our study. However as can be seen in Additional file 6: Table S5 the IHM did not differ between those T2DM patients who underwent or not valve surgery (22.6% vs. 20.49% respectively; p = 0.304). Chu et al. , in a prospective cohort of consecutively enrolled patients with definitive IE from 29 centers in 16 countries found that after multivariable analysis suffering DM was a negative predictor for surgical treatment for IE (OR = 0.52 [0.39, 0.69]). Other authors agree with us finding a lower surgical intervention rate in patients with DM .
In our investigation the mean length of hospital stay was around 27 days for those with and without T2DM. These values are much higher than those reported in investigations conducted in US [4, 33]. Recently Morita et al. describe a median length of stay of 10 (IQR, 6–17) days for those who survived to hospital discharge after IE during the period 2010–2014. Our equivalent figures for those who survived were over 28 days. From 1999 to 2010, using the Medicare inpatient Standard Analytic Files, Bikdeli et al.  reported that the mean length of stay for hospitalizations for endocarditis among patients aged 65 years of over consistently declined, from 9.6 (SD 11.5) days to 8.4 (SD: 8.9) days (p < 0.001). However they reported that around one quarter of patients was discharged to an intermediate care facility/skilled nursing facility, and another 10–17% to home health care. In Spain most patients (over 80%) are discharged to their home without any healthcare assistance and this may partly explain why the stay at the hospital is longer in our country. A report from Italy showed that median length of stay excluding hospital transfers increased from 23 to 25 days from 2000 to 2008, when hospital transfers were included the figures were 30 and 35 days respectively . In Denmark the median admission length was 31 days (25 and 75 percentiles: 13–45 days) . These last two countries have similar health service organization than Spain. Finally, in Brazil over a total of 203 consecutive patients admitted to a single tertiary care hospital between September 2005 and April 2017 the median duration of hospital stay was 42 days, ranging from 1 to 179 days .
Length of hospital stay is high for this condition, albeit it has been reduced somewhat in recent years. Perhaps, a faster identification of bacteremia with novel methods, the urgent notification of these positive microbiological results, a greater awareness on behalf of clinicians to proceed with an early transesophageal echocardiography at the detection of certain types of bacteremia , may be responsible for the observed reduction in the LOHS. Newly established strategies, such as sequential oral treatment for IE due to several types of microorganisms in patients with uncomplicated courses will probably allow even shorter regimens of intravenous treatment and lower hospital stays .
It is worth pointing out that mean cost per patient was consistently lower in people with T2DM than in people without diabetes. We do not have a straight explanation for this. The lower rates of heart valve surgery in the people with T2DM [24, 25, 44], or significantly shorter LOHS among the people with T2DM who died may partially account for this finding, but these arguments are merely speculative, since we lack the tools needed to provide a convincing reason to support this finding.
We were not surprised to find that hospital admission for IE had an overall high mortality (≈ 20%), although IHM decreased significantly over time. This figure is very similar to the rates described in the literature . Except for occasional reports , most studies have also shown improving outcomes in recent years.
In our opinion the larger decrease in the crude IHM found among those with T2DM than among non-diabetic patients (Table 1) may be a consequence of the improvement in the management and pharmacological treatment of T2DM patients that has been described in Spain over the last years .
We found female sex to be an independent risk factor for IHM in people with T2DM admitted for IE. In our environment, previous research has linked this association to gender-related differences in the rates of heart valve surgery , but rates of surgery did not differ between men and women in our study (data not shown). Other authors have claimed that the differences between genders regarding outcomes in IE most likely result from dissimilarities in coexisting medical conditions . Notwithstanding the importance of this topic, we cannot give a rational explanation for this finding.
After multivariable analysis valve surgery was not associated with IHM beside diabetes status. A previous review has reported that early valve surgery improve the outcome of patients with IE . Unfortunately, with our database it is not possible to know the time until surgery was conducted so the existence of a bias with this variable cannot be discarded.
T2DM was not associated with a higher IHM in our multivariable analysis. In the literature, both opposite and similar results have been published [13, 22,23,24,25,26,27,28,29,30,31].
Lin et al.  in a single tertiary care hospital showed that after adjusting the other confounding factors patients with DM had 3.29 times greater IHM rate compared to patients without DM. Kourany et al.  found similar results after multivariable analysis of 1055 people with IE (150 with DM) where DM was an independent predictor of mortality (OR 1.71, 95% CI 1.08–2.70), with crude IHM figures of 30.3% vs 18.6%, p = 0.001).
Some reasons have been suggested for the worse outcome in DM patients including: longer time from admission to the diagnosis of IE in the DM population which would resulted in the delay of suitable antibiotics treatment of and/or surgical intervention and the greater prevalence of S. aureus infection that is associated with higher rates of both complications and mortality in IE [13, 22, 24, 25].
However other studies didn’t identify DM as a risk factor [9, 23, 26, 28,29,30,31]. Fedeli et al. and Sy et al. in population based studies conducted in Veneto Region (North-Eastern Italy) and Australia did not find that DM was a risk factor for mortality [9, 28]. A previous Spanish investigation based in three tertiary care centers reported that multivariable analysis showed that DM had an independent association with development of septic shock (OR 2.282; 95% CI 1.186–4.393), but it was not a predictor of IHM . In France among 4405 IE patients admitted to ICU, 14% suffering DM, this condition was not a predictor of IHM .
Unfortunately, data on glycemic control (glucose levels, glycated hemoglobin) are not available in our database, and we cannot evaluate their impact on patient outcome. The only possible way to assess glycemic control in the SHDD is using ICD-9-CM codes 250.x0 and 250.x2, indicating controlled DM and uncontrolled DM, respectively. Doing this we found that 3320 (96.62%) have 250.x0 and only 116 (3.38%) have 250.x2 with the IHM being similar for both groups (688/3320; 20.72% vs. 26/116; 22.41%; p = 0.66). The extremely low prevalence of code 250.x2 makes in our opinion this method not useful to assess glycemic control.
It is known that the strict glycemic control can improve the cellular immunity and as a consequence the incidence and clinical prognosis of patients with DM when affected by critical cardiovascular illness [11, 18, 22, 24]. Critchley et al. describe that the largest relative associations between the poorest level of glycemic control (HbA1c < 11%) and optimal control (6–7%) were seen for bone and joint infections (IRR 8.71), endocarditis (IRR 5.56), and sepsis (IRR 3.64). However, the highest attributable risk fraction, defined as the percentage of infections that would not have occurred if all individuals had the same infection risk as those in the optimal control group of HbA1c (6–7%), was observed for endocarditis (26.2%) .
Wei et al. compared the IHM for three groups of patients (normoglycaemia, prediabetes and diabetes, according to the American Diabetes Association) and suffering IE: The IHM found were 3.4%, for normoglycaemia patients, 12.6% for those with prediabetes and 17.9%, for the diabetes groups (p < 0.001). Compared with the normoglycaemia group, the adjusted OR for IHM was 2.42 (95% CI 1.11–5.31) for prediabetes and 3.39 (95% CI 1.48–7.80) for the diabetes group . A larger, prospective, and more detailed (DM therapy, glycemic control, etc.) study is needed.
The strength and the novelty of our investigation are justified by the following reasons. First, we have analyzed data from an entire country over a 15-year period, providing data from 3436 T2DM patients suffering IE, this is one of the largest sample size ever described. Secondly, we use a PSM to improve the analysis of data, to our knowledge this method has not been used before for this topic. Finally, we focus specifically on T2DM patients providing the incidence for this patients as well as the evolution in their clinical characteristic and hospital outcomes and we identify the risk factors for IHM among this population. Nevertheless, beside those previously commented we should point out several limitations. Our data source was the SNHDD, an administrative database that relies on the information that physicians include in the discharge report and on manual coding on behalf of administrative staff. To our knowledge the ICD-9 codes for IE in the SNHDD have not been validated so far. However, results from three previous studies conducted in other countries, using ICD-9 codes in hospital discharge databases, suggested good accuracy for detection of endocarditis cases with reference to the revised Duke criteria [8, 28, 51]. The validity of the diabetes diagnosis in the SNHDD has been assessed in two previous studies, revealing a sensitivity of 55% and 63.7% and a specificity of approximately 97% [52, 53]. In a recent review of 12 studies (8 in US and 4 in Canada) using ICD-9 codes, specificity always showed high values (88% to 100%) with worse figures for sensitivity (26.9% to 100%) and the Kappa concordance index ranged from 0.6 to 0.9 . We lack longitudinal information about diabetes and other variables. Another limitation of our investigation is that only up to around 70% had identified causative pathogens. As the SNHDD is anonymized we cannot confirm that a patient that suffered a previous IE hospitalization and who moved from one hospital to another could be counted twice. Also is possible that patients with IE in the second diagnosis position aren’t admitted for IE but for another disease and have history of IE. The number of patients that had IE as a second diagnosis was 3206 (19.3%) with similar values beside T2DM status (19.05% among T2DM and 19.34% among non-diabetic patients). In both groups over 90% of patients with a second diagnosis of IE the primary diagnosis were ICD-9 code 996.91 (Infection and inflammatory reaction due to cardiac device, implant, and graft” or valve disease ICD-9 codes (424.0, 424.1, 424.2, 424.3). Therefore we consider this error improbable and if exists of a very small magnitude that would not affect the main conclusions of our investigation as it affect similarly those with and without diabetes. HIV patients or those with drugs abuse suffering IE were excluded for two reasons, first because the clinical characteristics of IE among may be different to those without these conditions and secondly this is a very infrequent condition among T2DM patients (only 16/3436 [0.46%] vs. 415/13,190 [3.15%] among non-T2DM sufferers). Furthermore, we cannot rule out that coding practices may have changed somewhat over time. Finally, residual confounding not accounted for could be influencing the results of the multivariable analysis of the factors associated with in-hospital mortality.