Do Stroke Care Processes Modify Clinical Outcomes in Patients with Stroke-Associated Pneumonia? A Registry Cohort Study in England and Wales

Stroke-associated pneumonia (SAP) occurs in 8–13% of people hospitalized with stroke [1, 2] Previous studies have shown that SAP increases mortality, in-hospital stay, and healthcare costs [3‒5]. However, whilst the majority of these studies have adjusted for potential confounding effects of baseline patient characteristics on admission [3, 6‒8], the role of acute stroke care processes and their timing when considering outcomes from SAP in a real-world setting has not been considered. Whilst stroke unit care reduces the incidence of SAP compared to non-stroke unit settings, the effects on outcomes of SAP are not well characterized [9]. Several studies have explored/investigated associations between stroke care processes (e.g., dysphagia screening, swallowing dysfunction treatment, feeding restrictions) and outcomes of SAP, but the role of stroke care processes in outcomes from SAP requires further study [10‒12].

Exploring the potential confounding of stroke care processes on outcomes in SAP is important as it could have implications for delivery of stroke unit care, quality improvement initiatives, or design of intervention trials. The main aim of this study was to investigate the association of SAP with clinical outcomes in patients who remained in hospital after 7 days, whilst exploring the influence of the confounding effects of baseline clinical characteristics and markers of acute stroke care.

The Sentinel Stroke National Audit Programme (SSNAP) is a mandatory registry of stroke patients entering hospitals in England, Wales, and Northern Ireland designed to gather baseline characteristics, demographics, and stroke care measures. It falls under the purview of the Healthcare Quality Improvement Partnership (HQIP), which in turn controls access to the data. HQIP approved the final data transfer. All data access requests must be made directly to HQIP. Ethical approval was not necessary as the SSNAP gathers patient data with their consent. For further information, contact the SSNAP directly.

The SSNAP data here formed an observational cohort of patient level data for all confirmed strokes from April 1^st 2013 to December 31^st 2018 in England and Wales. A stroke patient was defined as having SAP if they were recorded as having a new antibiotic initiation for suspected pneumonia within the first 7 days of admission. We excluded patients who were discharged or died in the first 7 days in order to remove the differential time component of the definition of antibiotic initiation in the two comparison groups. Patients with greater than 365 days or stroke units with less than 150 admissions per year were also excluded. This was in order to improve the representativeness of the follow-up period in the two comparison groups and the stroke units included in the study, i.e., removal of those stroke units that had been repurposed for a small population of stroke patients [13]. Figure 1 describes how we arrived to our sample population.

The clinical outcomes were the modified Rankin Scale (mRS) score on discharge from hospital, total length of in-hospital stay until discharge from hospital, and in-hospital mortality.

Patient and clinical characteristics associated with worse clinical outcomes and with SAP in previous studies were extracted. These included patient age, sex, National Institute of Health Stroke Scale (NIHSS) score on admission, premorbid mRS on arrival, and the presence of other comorbidities such as congestive heart failure, diabetes, atrial fibrillation, hypertension, and previous stroke or transient ischaemic attack [13, 14].

In addition to clinical characteristics, we also adjusted for several markers of stroke care thought to be associated with both health outcomes and the development of SAP. These were based on the SSNAP’s markers of good quality stroke care [15], and hyperacute or acute processes were selected as SAP occurs most commonly in the first 72 h [1, 16]. The selected markers were arrival at a stroke unit within 4 h from the symptom onset (Yes/No); received a swallow screen within 4 h from admission (Yes/No); assessment by a physiotherapist, speech and language therapist, and an occupational therapist within 72 h from admission (all in a single composite binary score – Yes – was seen by all therapists in 72 h, No – was not seen by all therapists in 72 h); and if the patient, if eligible, received thrombolysis (Yes – patient received thrombolysis, No – the patient did not receive thrombolysis).

This SSNAP dataset was representative of the hierarchical structure of stroke care, where stroke patients were nested within stroke units. To account for the within versus between stroke unit correlation, a multilevel mixed model approach was selected where the stroke unit was set as the random intercept. Length of in-hospital stay and the mRS score on discharge were modelled using a multilevel negative binomial model and ordinal logistic model, respectively. Time (in days) to in-hospital mortality post-7 days from admission was modelled using a multilevel parametric survival analysis model using the exponential distribution. This model was chosen over the proportional hazard function as we believed a priori that the proportional hazards assumption would not hold over the 365-day follow-up. A Kaplan-Meier curve described the time to in-patient mortality between SAP and non-SAP patients after 7 days. Appropriate effect estimates (and 95% confidence intervals [CIs]) in the form of incidence rate ratio, odds ratios and the hazard ratios are reported. The clinical outcomes were modelled in three stages to understand the influence of patient and clinical characteristics and stroke care markers on clinical outcomes. The first stage modelled SAP versus non-SAP as a single fixed-effect covariate along with the stroke unit random intercept. In stage two, we added the patient and clinical characteristics as fixed effects covariates, and finally in stage three, stroke care markers were added to the model.

The initial dataset comprised of 458, 829 patients across 328 stroke units, of which 39,467 (8.6%) were diagnosed with SAP. A total of 5,304 (1.2%) patients had missing data regarding SAP status. After applying the exclusion criteria, the dataset comprised of 201,778 patients across 169 stroke units (Fig. 1). SAP was present in 28,688 (14.2%) patients, with 2,217 (1.1%) patients having missing data regarding SAP status who were consequently excluded. The median length of stay for SAP patients was 24 days (IQR 13–18) and in non-SAP patients was 18 days (IQR 10–35). There was a higher proportion of SAP patients with an increased mRS score on discharge (Fig. 2) compared to non-SAP patients, with 3,191 (7.3%) having an mRS of 3; 5,992 (12.8%) having an mRS of 4; 5,664 (23.4%) having an mRS of 5; and 11,425 (32.9%) having an mRS of 6. The median time to in-hospital death for SAP patients was 15 days (IQR 9–26), and the median for non-SAP patients was 15 (IQR 9–28). A full description of clinical characteristics and clinical outcomes can be found in Tables 1and2.

In patients with SAP remaining in hospital after 7 days and after adjusting for clinical characteristics and stroke care processes, there was an increased length of stay with an incidence rate ratio of 1.27 (95% CI: 1.25–1.30). In patients surviving to at least 7 days, SAP was associated with increased disability on discharge, with an odds ratio of 2.9 (95% CI: 2.9–3.0). Finally, patients with SAP surviving to at least day 7 after admission had an increased risk of in-patient mortality (hazard ratio 1.78 [95% CI: 1.74–1.82)]. Figure 3 presents a Kaplan-Meier curve describing the mortality association. Results of the analysis of each outcome are presented in Table 3. Full results are available in online supplementary Table S1–S3 (for all online suppl. material, see www.karger.com/doi/10.1159/000524917).

The real-world impact of modern, interdisciplinary stroke unit care on morbidity and mortality from complications of stroke is uncertain. However, better understanding the relationships between care processes and their timing with clinical outcomes of SAP could drive further quality improvement and research efforts to reduce the adverse impact of this common complication. In a large national registry representative of UK stroke unit care, we found that SAP was associated with worse clinical outcomes in stroke patients despite adjustment for acute care processes and their timing. These findings suggest that the acute stroke care processes we investigated did not materially alter the observed associations between SAP and poor clinical outcomes.

Compared to previous studies [3, 14, 17], we only included patients who remained in hospital after the initial 7-day exposure period for SAP. By using this approach, we accounted for the potential misclassification bias in SAP diagnosis itself and between our comparison groups related to the time-dependent nature of SAP diagnosis in our data. By using this approach, the possible masking caused by stroke severity on clinical outcomes is diminished. This is because milder strokes could have been discharged before 7 days, and severe strokes could have died within the first 7 days, in both cases before SAP was diagnosed. Our approach accounts for this potential influence and is more representative of the association between SAP and long-term clinical outcomes occurring in hospital.

A possible explanation for our findings is that the care processes used in the analyses are not necessarily focused on SAP but rather facilitate rapid stroke care and delivery to improve outcomes more generally or by facilitating speed of delivery of particular interventions such as thrombolysis or arrival on the stroke unit. This effect can be seen in the reduction of worse clinical outcomes in each of the included stroke care processes included in our models. Another possibility is that these processes could be associated with early discharge and have limited impact on those with longer term in-patient health outcomes. Limited granularity of the data on stroke care processes may have contributed to the presence of residual confounding and limited our interpretation of their influence. Whilst our findings compound on existing evidence for good quality stoke care and outcomes [18], there are certain aspects of stroke care that need to be analysed further, such as the amount of physiotherapy a patient receives, what clinical care decisions were made such as positioning or nasogastric feeding, or what type of specific treatment each patient received (including antibiotics). Analysing these variables could add further granularity to the existing evidence surrounding stroke care.

Our findings have possible implications for clinical practice. Currently, there are no validated guidelines specifically addressing prevention and management of SAP. There have also been several clinical trials and observational studies in selected populations looking at different preventative measures which can reduce the risk of SAP development such as preventive antibiotics; however, they have no effect on clinical outcomes [19‒21]. It is also important to note that care processes may be associated with the development of SAP, but it is also vital to know whether they reduce poor outcomes in SAP. Our data suggest that acute care processes were not associated with outcomes from SAP, highlighting the need to focus research efforts on preventative strategies and effective SAP treatment, such as identifying the antibiotics classes and duration that could benefit SAP patients the most [22, 23]. The preventative strategies that could be explored further include oral care, chest physiotherapy, and metoclopramide [20, 24, 25].

Our study has several strengths. These include the large sample size of national data with a high-case ascertainment. The data are representative of the UK patient pathway, across multiple stroke units over several years. More detailed measures of health outcomes provides more robust findings, compared to previous work that has focused on simpler binary outcomes [3, 17]. We opted to use ordinal outcome data or that based on time to the outcome. Our study also has several limitations. We were limited in terms of the recorded variables available to be included in the models as they were limited by the data that are recorded by the SSNAP. Several unmeasured factors that have the potential to contribute to confounding were not available and could be included in future studies. These include baseline characteristics such as preceding dementia or chronic chest disease, stroke subtype, nursing interventions such as positioning, details of treatment received such as physiotherapy and therapy duration, severity of SAP, microbiological aetiology, and antibiotics received. We were also unable to account for other possible infections or complications that could also have been associated with the clinical outcomes in our study. We were also unable to account for organizational level confounding, such as nursing staff or therapist levels or stroke unit capacity, which would likely be associated with SAP and influence clinical outcomes. We were also limited regarding the detail of mortality as we did not have access to the cause of death or if SAP had an association with death after discharge. Finally, it is uncertain how generalizable our findings are outside of UK stroke unit care, despite some overlap with measures of good stroke unit care, and require evaluation in alternative healthcare settings.

Our findings indicate that SAP continues to be associated with worse clinical outcomes in stroke patients, even after adjustment for baseline prognostic factors and acute care processes. Our findings highlight the need to focus on preventive strategies and management of SAP in order to improve clinical outcomes.

We would like to thank Kaili Stanley, Sabrina Ralph, and Professor Martin James at the SSNAP, as well as all the hospital SSNAP teams and patients involved in SSNAP.

The data presented in this study were provided by the SSNAP and HQIP. The study was approved by the data provider, approval number HQIP DARF 296. This study did not require additional local ethical approval in accordance with local/national guidelines. Written informed consent from participants was not required for the study presented in this article in accordance with local/national guidelines.

The authors have no conflicts of interests.

No funding was used in this study.

Marco Antonio Lobo Chaves submitted the data request and contributed to the data analysis and the first draft. Craig J. Smith, Andy Vail, and Benjamin Bray designed the study. Matthew Gittins contributed to the data analysis. All the authors contributed to the editing of the final draft.

All data enquiries should be directed to HQIP and SSNAP (https://www.hqip.org.uk/).