This observational, retrospective multicenter study was conducted in four hospitals that comprise five different rehabilitation settings: two SICUs, two high-specialty PAUs, and one LTC facility. These wards are part of a clinical care pathway for sABI patients, encompassing the acute and rehabilitation phases up to discharge at home or to community facilities. All those settings are subjected to similar prevention and infection control measures according to the clinical care pathway. The two SICUs interact with the intensive care unit and neurosurgery to ensure a timely neurorehabilitation approach. They define the diagnosis, begin the rehabilitation process, and once patients are stabilized, they are directed to the next level of care. The two PAUs provide comprehensive care for sABI patients, offering separate units for disorders of consciousness with a long-term rehabilitation process. Based on the clinical stability and recovery of the patients, the trajectories of the rehabilitation path could differ. The possible trajectories were only SICU → PAU (N = 49), SICU → LTC (N = 2), SICU → PAU → LTC (N = 1), and PAU → LTC (N = 1). Unique trajectories were only SICU (N = 22), PAU (N = 39), and LTC (N = 20).

Patients not eligible for intensive rehabilitation treatment are directed to the LTC (13). All five wards use the same diagnostic and treatment approach with an individually customized rehabilitation plan that also involves the patients' families.

The study was approved by the local Ethics Committee (protocol No. 609-2019-OSS-AUSLIM) and did not receive funding. Informed consent was obtained when possible or waived in accordance with the General Authorization of the Privacy Guarantor No. 09/2016 on observational retrospective studies.

Data on patients hospitalized in 2018, including patients admitted in 2017 and discharged by 31 December 2018, were extracted from medical records.

Data extraction was made by experienced clinicians, including some of the authors (GBC, AB, and EB), who entered the data extracted in a structured Excel file, using agreed criteria on case definitions of healthcare-associated infections based on ECDC Codebook (14) and are summarized in Supplementary Table S1.

During the study, some patients were admitted to more than one setting. In these cases, data were collected from both settings. Inclusion criteria were as follows: age ≥18 years, sABI of any etiological origin (traumatic, non-traumatic hemorrhagic, anoxic, infective, neoplastic, or toxic-metabolic), Glasgow Coma Scale ≤ 8 for at least 24 h, and impairments of physical, neurocognitive, and/or psychological function that involve a severe disability.

The number and etiology of infections occurring during hospitalization were collected. The diagnostic criteria of HAI were based on the ECDC Codebook (14) and are summarized in Supplementary Table S1. The most common infections are bloodstream infections (BSIs), UTIs, and pneumonia.

Other infections were considered if reported in medical records. They included Clostridium difficile infection, otitis, cellulitis, ventriculitis, phlebitis, epididymo-orchitis, skin and soft tissue infections, and intra-abdominal infections.

We omitted fewer common infections.

Infections were classified as non-MDR HAI or MDR HAI according to the organism found and based on the international expert proposal of Magiorako et al. (10) study.

The primary outcome was the incidence of HAIs and MDR HAIs. Secondary outcomes included LOS and functional recovery measured using the Disability Rating Scale (DRS) (15) and the Level of Cognitive Functioning Scale (LCF) (16), which are routinely administered in SICU and PAU settings. These functional assessment scales were not administered in the LTC setting.

LCF is a well-established tool to assess cognitive functioning in post-coma patients, validated in an Italian study (17). Patients are classified into eight levels, from 1 (non-responders) to 8 (purposeful-appropriate person); the higher the value, the better the cognitive function.

DRS is a 30-point scale measuring eight areas of functioning: eye-opening, verbalization, motor response, level of cognitive ability for daily activities of feeding, toileting, and grooming, overall level of dependence, and employability (18). Scores in each area (rated from 0 to 3 or 0 to 4 or 0 to 5) are summed to yield a total score between 0 and 30, with a higher score denoting lower functioning.

The predictors of HAIs were classified into pre-admission characteristics, inpatient indicators, clinical parameters, and device-related data. Pre-admission variables were socio-demographic characteristics, history of previous diseases, type of injury, and any previous surgery. We added the condition of associated trauma for traumatic patients (Table 1) for those who suffered other traumatic lesions such as thoracic, skeletal, abdominal, or spinal cord. We also collected comorbidities, meaning specifically the presence of coexisting or additional pathologies with respect to brain damage with an infectious impact.

Inpatient indicators included the time between injury and admission, any transfers to other departments, any surgery during hospitalization, and any ward shifts caused by clinical worsening.

Clinical parameters were CPE colonization on admission or during rehabilitation and nutritional parameters (protein, albumin, and prealbumin assay) on admission. Moreover, clinical conditions with a potential negative impact on the functional outcome not directly related to HAIs (cardiovascular complications, osteoarticular problems, wound onset, and dystonia) were collected.

Devices included an indwelling urinary catheter, a central venous catheter, a tracheotomy tube, and a percutaneous endoscopic gastrostomy. The number of medical devices used for each patient was collected.

Continuous variables were summarized using mean and standard deviation (±SD) when normally distributed, and median and interquartile range [IQR] otherwise; categorical variables were summarized using frequencies.

The incidence of HAIs and MDR HAIs per 100 person-days was calculated using the length of hospital stay as the exposure time. HAI incidence was compared between the three settings. In case of a significant difference, the rehabilitation setting was included as an adjustment variable in the subsequent analyses.

A negative binomial regression analysis was used to identify the risk factors associated with HAI and MDR HAI incidence. Significant variables were included in a multivariable model. The results were reported as an incidence rate ratio (IRR) and 95% confidence interval (95% CI). All the significance levels reported refer to comparisons of regression coefficients in the Poisson and negative binomial models. When more than two groups were compared, a Bonferroni correction to the probability level was applied. The adjustment for the rehabilitation setting is obtained by including the rehabilitation setting in the model as two dummy variables, one for PAU and one for LTC, and using SICU as the reference category.

In a secondary analysis, including data from SICU and PAU settings only, patients were classified into three mutually exclusive groups: no infection, at least one HAI (non-MDR), and at least one MDR HAI.

These groups were compared on LOS and the rehabilitation outcomes (DRS and LCF scores at discharge), using negative binomial regression and linear regression, respectively. The results on LOS were reported as predicted LOS (in weeks) with 95%CI. In linear regression analysis, a model adjusted for rehabilitation setting and functional score on admission was initially estimated, then a multivariable model adjusted for other factors was obtained. Regression coefficients were reported with 95%CI.

The backward stepwise variable selection procedure was applied to obtain parsimonious multivariable models (p for removal=0.05). Robust standard errors were estimated using a clustered sandwich estimator to take into account repeated measurements (i.e., admissions in different settings) on the same individual. Bonferroni's correction was applied for multiple comparisons. The Bonferroni correction consists of adopting a significance level adjusted for the number of comparisons when the groups compared are more than 2.

In this case, the probability level was adjusted for three pairwise comparisons among settings (0.05/3 = 0.017). The statistical software (Stata), when the Bonferroni correction is requested, provides already adjusted p-values. For instance, if the p-value of a test is 0.002, it is provided in the output as 0.002^*3 = 0.006. All the p-values in our results are adjusted p-values when they refer to comparisons of the three settings.

Statistical analyses were carried out using Stata version 15 (StataCorp, College Station, TX). The significance level was set to a p-value of < 0.05.

The analysis included 188 records of 134 patients admitted to at least one of the four hospitals. During the study period, 49 patients were transferred from SICU to PAU, 2 patients from SICU to LTC, 1 patient from PAU to LTC, and 1 patient crossed all three settings (Supplementary Figure S1). Overall, the analysis included the data of 90 patients hospitalized in SICU, 74 patients hospitalized in PAU, and 24 patients hospitalized in LTC. Patients' characteristics by setting are reported in Table 1. The mean age was 53.3 ±18.2 years, and 66.5% of patients were male. Traumatic etiology accounted for 40.4% of cases, and 23.9% of patients had comorbid conditions. In LTC, the sample was older on average, had more comorbidities, and had a lower frequency of traumatic etiology. The median LOS was 21 days [13–42] in SICU, 103 days [55–203] in PAU, and 48 days [9–169] in LTC.

The HAI incidence ranged from 1.2 per 100 person-days in PAU to 1.9 in LTC and up to 3.7 in SICU (Table 2). The incidence in SICU was significantly higher than in the other settings (p < 0.001 SICU vs. PAU; p = 0.034 SICU vs. LTC). MDR HAI incidence was 0.3 per 100 person-days in PAU, 0.5 in LTC, and 1.3 in SICU. The incidence in SICU was significantly higher compared to the other two settings (p < 0.001 SICU vs. PAU; p = 0.026 SICU vs. LTC).

Supplementary Figure S2 reports the HAI and MDR HAI incidence in each rehabilitation setting, according to the infection etiology. Overall and MDR bloodstream infections (BSIs) had a significantly higher incidence in the SICU as compared to the other settings. The overall UTI incidence was higher in SICU (0.68, 95%CI: 0.33–1.03) than in LTC (0.45, 95%CI: 0.11–0.78) and PAU (0.20, 95%CI: 0.10–0.31), but only the difference between SICU and PAU was significant (p = 0.017). On the other hand, the overall incidence of pneumonia was similar in SICU (1.01, 95%CI: 0.46–1.55) and LTC (0.94, 95%CI: 0.37–1.52), while it was significantly lower in PAU (0.36, 95%CI: 0.21–0.52). MDR pneumonia incidence had a similar trend. A low incidence of MDR UTIs, skin and soft tissue infections, and intra-abdominal infections was observed in all three settings. The incidence of other infections was higher in SICU as compared to the other two settings. Overall, 280 infections occurred; the most frequent pathogens involved were P. aeruginosa (15.7%), K. pneumoniae (14.3%), E. coli (10%), P. mirabilis (7.4%), Candida spp. (4.8%), E. faecalis (4.3%), and A. baumannii (3.9%). In 26.1% of cases, the pathogen was not identified.

Among the 280 HAIs, 85 were identified as MDR (30.3%). The MDR frequency on certain isolates was higher than 50%, specifically: K. pneumoniae (55%, N = 33), A. baumannii (89%, N = 9), methicillin-resistant S. epidermidis, MRSE (N = 8), methicillin-resistant S. aureus, MRSA (N = 6), and E. cloacae (100%, N = 4).

Table 3 shows the factors associated with HAI and MDR HAI incidence. In the regression analysis adjusting for rehabilitation setting only, older age, any comorbidities, lower LCF and higher DRS scores upon admission, higher number of devices, and CPE colonization were all significantly associated with HAI development during inpatient rehabilitation. Conversely, a high prealbumin plasma value on admission was associated with a decreased likelihood of developing HAI. In the multiple regression analysis, older age, an increased number of devices, and CPE colonization remained significant. Older age and CPE colonization were also significant risk factors for MDR HAI incidence.

In the SICU setting, LOS was significantly higher in the HAI group compared with the no-infection group (4.7 vs. 2.6 weeks, p = 0.008) and even higher in the MDR HAI group (10.2 weeks, p = 0.003) (Figure 1A). In the PAU setting, LOS was on average 11 weeks in the no-infection group and approximately 25–26 weeks in the HAI and MDR HAI groups. After adjusting for the number of devices, surgery during rehabilitation, and complications, the differences between the three groups remained significant (Figure 1B).

In a regression model adjusted for admission score and rehabilitation setting, being in the MDR HAIs group was significantly associated with higher DRS and lower LCF scores at discharge (Table 4). However, these associations were no longer significant after adjusting for the number of devices, surgery during rehabilitation, and complications. Notably, in the presence of CPE colonization and complications, disability at discharge increased as the time from injury to hospitalization increased. Similarly, the LCF score at discharge was lower for patients with an increased time from injury to hospitalization, patients who underwent neurosurgery before rehabilitation, and those with an etiology other than traumatic or vascular.