The role of the Sapienza GLObal Bedside Evaluation of Swallowing after Stroke (GLOBE-3S) in the prevention of stroke-associated pneumonia (SAP)

We conducted a case–control study of acute stroke patients detected with dysphagia. In the case group, we recruited 50 patients who had been tested for dysphagia with GLOBE-3S in our previous prospective trial (for methodological details, see elsewhere [19]). As a control group, we retrospectively recruited 50 patients matched by age, sex, and stroke severity assessed with National Institute of Health Stroke Scale (NIHSS) out of 275 patients from a previously described cohort of acute stroke patients [21] who were screened for dysphagia with the water swallow test (WST), as per routine in our ward at that time. All patients underwent the swallowing screening within 24 h from admission (but anyway within 72 h from stroke onset).

The matching between groups was performed by the study coordinator, who was blinded to all patients’ information except those used for matching. Patients were paired by using a custom-made approach in which each patient of the case group was paired with the first control patient having the same value (± 1) in the variables of interest (age, sex, and NIHSS). In the second permutation, each case–patient was paired with the second patient in the control list having the same characteristics. About ten permutations filled all case–patients, and the best one in terms of similar average and standard deviation was chosen. An independent neurologist, blinded for group allocation (i.e., patients’ dysphagia detection method), accurately collected clinical data from patients’ clinical records that were used to perform group comparison.

The diagnosis of SAP was performed in both groups according to the 2015 Consensus Group criteria by a neurologist blinded to the swallowing assessment [22] within the 2 weeks from stroke onset. Only patients who failed either the GLOBE-3S or the WST test were defined as having dysphagia.

The study and all experimental protocols were conducted according to the Declaration of Helsinki and approved by the ethics committee of “Sapienza” University of Rome.

Inclusion criteria for recruiting patients were for both groups: age ≥ 18, evidence of an ischemic or hemorrhagic stroke at magnetic resonance imaging (MRI) or computed tomography (CT), symptoms onset within 72 h, and the possibility to give informed consent. Informed consent was obtained from all subjects. We excluded patients with a transient ischemic attack (TIA), subarachnoid hemorrhage, cerebral sinus venous thrombosis, Glasgow Coma scale score < 13 [23], and those patients with a history of previous swallowing impairment or a medical condition that may affect swallowing function. We also excluded all patients with pneumonia or any infection signs upon admission.

Upon admission, all patients underwent a standard neurological clinical examination. Stroke diagnosis was confirmed by MRI or CT scan. Stroke severity was assessed using the NIHSS [24].

Details about GLOBE-3S were presented in previous studies from our group (CIT) [19]. In brief, the GLOBE-3S test consisted of a combination of Toronto Bedside Swallowing Screening Test (TOR-BSST©), pulse oximetry, and laryngeal elevation assessment. Patients were assessed for dysphagia using the TOR-BSST© [25]. As regards pulse oximetry, oxygen desaturation was monitored during and 120 s after TOR-BSST©’s first part (10-ml water intake) examination. A pulse oximetry decrease greater than or equal to 2% was considered a pathological result [19]. Laryngeal elevation was also monitored while performing TOR-BSST©’s first part (starting position and final position were assessed at the beginning and end of this part, respectively) and according to the literature [26], a laryngeal elevation < 2 cm was considered abnormal. Those patients who failed in any of these three items were classified as having dysphagia.

The water swallow test (WST) consisted of consecutive swallowing trials of different water volumes. Namely, this step-by-step WST was a two-stage screening performed with 5 and 60 ml according to Smithard et. al [20]. Patients with choking, coughing, or voice change were considered positive for dysphagia. In our stroke unit, WST was part of the standard protocol for the detection of dysphagia [21] before we implemented the GLOBE-3S.

In both groups, all patients diagnosed with dysphagia were allocated to early enteral nutrition via nasogastric tube (NGT) and referred for the speech and language pathologist (SLP) management according to the evidence in the literature on the management of dysphagia in acute care settings [27‐29]. This besides consisted of compensatory approaches, including modification of fluid and food consistencies; postural techniques, such as adopting a chin tuck position; swallow strategies, such as a supraglottic swallow to help patients in reducing aspiration; and exercises aimed to strengthen the musculature. Patients were also hydrated via an intravenous infusion of 0.9% saline solution and started on NGT within 24–48 h. Gastric residual volume was regularly controlled (i.e., every 6 h).

Our aim was to assess if the 100% of sensitivity obtained by the GLOBE-3S, which entails that all patients with dysphagia were rapidly identified, may lower the risk of developing pneumonia during the acute phase of stroke. Estimating the required sample size for this aim may be complicated due to the unknown estimated effect size, namely, SAP mechanisms are multifactorial and overlapping so that the exact percentage of SAP exclusively due to dysphagia (i.e., those SAP that could be prevented with a highly sensitive screening for dysphagia) is unknown. Anyway, we conducted a priori sample size calculation assuming that dysphagia is directly responsible for almost 50% of SAP [11]. Giving the 100% sensitivity and the 100% negative predictive value (NPV) achieved by the GLOBE-3S in the validation study (i.e., no false-negative results) [19], it is acceptable to expect up to 50% of SAP reduction (i.e., to expect avoiding all SAP due to dysphagia). Thus, considering that SAP incidence ranges from 9 to 38% across the studies [30] by means of t-test (a priori goodness of fit), we calculated that a total sample of 56 stroke patients (28 patients for each group) would provide a power of 95% to have the SAP incidence (α-error: 0.05, effect size: 0.99, two tails).

To compare the number of SAP between patients of the two groups (GLOBE-3S and WST) and between patients with and without dysphagia, we used chi-squared or Fisher’s exact test depending on the number of subjects per cell.

To predict the occurrence of SAP in the two groups, we used discriminant function analysis (DFA). DFA is a multivariate nonparametric technique whose aim is to maximize the classification of a pool of subjects into different groups according to an outcome variable. DFA uses a linear combination of contributions obtained by a set of predictive variables. We used the occurrence of SAP as an outcome measure (SAP vs. non-SAP). As predictors of the outcome, we used parameters from literature — known to be risk factors for the development of pneumonia [5, 21, 31] as well as the diagnostic method used to screen for dysphagia (e.g., GLOBE-3S vs. WST). We used age, sex, stroke type, NIHSS value, degree of leukoaraiosis, urinary tract infections, presence of swallowing disorders (without any distinction between WST and GLOBE-3S), and swallowing detection methods.

Classification of single cases was performed with the Mahalanobis distance (MD) method. Methodological and technical details on DFA can be found elsewhere [32‐34]. Based on the previous results, we used DFA to evaluate which parameters were the stronger predictors of SAP occurrence in our sample of patients.

Data were analyzed using SPSS software for Windows (version 22; IBM Corporation, New York, NY, USA) and STATISTICA (version 7, StatSoft, OK, USA).

When analyzing SAP incidence according to the detection of dysphagia (Fig. 2), patients diagnosed with dysphagia had a higher number of SAP compared to patients without dysphagia (27 vs. 7 patients respectively, p = 0.006).

In patients without dysphagia, pneumonia was diagnosed in none of the patients of the GLOBE-3S group, compared to 7 patients (31.82%) of the WST group (p = 0.02) (Fig. 2). Stroke severity was similar in the two groups (mean NIHSS score 5.75 in the GLOBE-3S group and 5.86 in the WST group, p = 0.94). For patients without dysphagia, the OR for pneumonia could not be calculated due to the lack of pneumonia in the group screened with GLOBE-3S (Fig. 2).

On the other hand, in patients diagnosed with dysphagia, pneumonia occurred in 10 patients (32.26%) of the GLOBE-3S group compared to 17 patients (60.7%) of the WST group (p = 0.04) (Fig. 2). As seen in non-dysphagic patients, also in this case, the stroke severity was similar between patients in the WST and GLOBE-3S. However, in the WST group, the NIHSS score was about 2.5 points higher although in a not significant manner (mean NIHSS score 12.29 in the WST group and 9.81 in the GLOBE-3S group, p = 0.14). For patients with dysphagia, an OR of 3.2 (95% CI: 1.1–9.4, p = 0.03) was found for the risk of pneumonia between the WST and GLOBE-3S methods (Fig. 2).

Overall, SAP was detected in 34 patients (28.9%): 24 in the WST group and 10 in the GLOBE-3S group. The within-group analysis between dysphagic and non-dysphagic patients in either GLOBE-3S or WST (Fig. 2) showed that the difference in pneumonia rate between patients with and without dysphagia resulted significant within both the GLOBE-3S group (0 vs. 32.26%, p = 0.01) and the WST group (31.82% vs. 60.71%, p = 0.04).

We used DFA to evaluate which parameters were the stronger predictors of SAP occurrence in our sample of patients (Table 2). Overall, the model showed a high prediction capability with a Wilk’s Lambda value of 0.66 (p < 0.0001).

The stroke severity, as evaluated by the NIHSS value, remains the major descriptor being significantly associated with SAP with a 0.9 degree (p = 0.007), followed by the diagnosis of swallowing disorder obtained by using the GLOBE-3S rather than the traditional WST method, which is associated with SAP with a 0.88 degree (p = 0.002). The other factors, such as urinary infections, gender, and age improved the ability of the model but were not significantly related to SAP individually. Interestingly, the detection of dysphagia per se did not appear in the model.

Furthermore, if we classify patients by using the statistical model, the predictions of those patients who are likely to develop pneumonia would be evenly distributed among the groups (Table 3).

GLOBE-3S patients without dysphagia were those better classified by the model: DFA predicted none of them to have SAP, and the datum coincides with the real number. The error between prediction and real data in the other groups was 6.4% for GLOBE-3S patients with dysphagia, 7.1% for WST patients with dysphagia, and 4.4% for WST patients without dysphagia. Overall, the model showed a percentage of correct predictions equal to 91%.

Since DFA analysis showed that the stroke severity remains the major descriptor being significantly associated with SAP with a 0.9 degree (p = 0.007) and considering that the NIHSS is also associated with post-stroke dysphagia [21, 35], those patients with moderate-to-severe stroke are somehow expected to develop SAP, and probably are the ones who less require a highly sensitive dysphagia screening for SAP to be prevented.

On the contrary, patients with mild stroke are the ones who are most likely to be missed from most of current dysphagia bedside screenings. Thus, assuming that a lower sensitive dysphagia screening method might misidentify dysphagic patients with mild stroke who could anyway develop SAP, we performed a subgroup analysis aimed to investigate the incidence of SAP in patients with mild stroke (i.e., patients with NIHSS ≤ 5) [36]. Overall, we found 33 patients with mild stroke: 15 in the GLOBE-3S group and 18 in the WST group. Out of 33 patients with mild stroke, only 5 patients developed SAP. All 5 patients were in the WST group, and out of these, 3 were screened as non-dysphagic, whereas 2 patients were correctly identified as dysphagic.

The main weakness of the present study was its observational design, as well as the lack of randomization. Anyway, in our study, the two groups of patients were recruited with the same inclusion criteria (e.g., enrollment within 72 h from stroke onset), in the same stroke unit, and then treated by the same stroke team. Patients of the WST group were blindly selected from a larger cohort of patients based on three parameters: sex, age, and stroke severity. Moreover, based on the clinical and radiological characteristics defined as predictors of dysphagia [5, 21, 31], it is worth noting that the two groups did not differ for all the other main covariances (e.g., type of stroke, leukoaraiosis, urinary infection), although these parameters were not matched a priori.

Three patients from the GLOBE-3S group were excluded since an admission chest radiograph showing no prior pneumonia was not available for these patients. Thus, considering the current guidelines for SAP [22], we had to exclude them from the study. Anyway, they did not report any sign of SAP, and therefore, their exclusion is unlikely to have biased the results.

Although DFA analysis showed that the detection method of dysphagia is an independent predictive factor for SAP, this is not a superiority study, and the absence of SAP in those patients who passed the GLOBE-3S screening may partly be due to the relatively small sample size. Then, future crossover randomized controlled trials, with a larger sample of patients, are needed to draw more consistent conclusions. Anyway, our results are in line with those from Joundi et al. [12] who found that only 1% of stroke patients were screened as non-dysphagic with the TOR-BSST (an accurate screening tool that is already part of the GLOBE-3S) developed SAP.