The Benefit of Dysphagia Screening in Adult Patients With Stroke: A Meta‐Analysis

Abstract

Background

Early identification of dysphagia aims to mitigate the risk of health consequences in adults poststroke; however, the evidence from experimental trials alone is inconclusive. This meta‐analysis assessed dysphagia screening benefit from both trial and observational data.

Methods and Results

Seven electronic databases were searched to December 2019. Unique abstracts and full articles were screened for eligibility by 2 independent blinded raters using a priori criteria and discrepancies resolved by consensus. Included studies were summarized descriptively and assessed for methodological quality using Cochrane Risk of Bias Tool. Across studies, pooled estimates of health benefit were derived for homogeneous data using Review Manger 5.3. From the yield of 8860 citations, 30 unique articles were selected: 24 observational and 6 randomized trials. Across studies, comparisons varied: no screening versus screening, late versus earlier screening, informal versus formal screening, pre‐ versus postscreening, and pre‐ versus poststroke guidelines that included screening. Pooled estimates across comparisons favored experimental groups for pneumonia odds ratio (OR), 0.57 (95% CI, 0.45–0.72), mortality OR, 0.52 (95% CI, 0.35–0.77), dependency OR, 0.54 (95% CI, 0.35–0.85), and length of stay standardized mean difference, −0.62 (95% CI, −1.05 to −0.20).

Conclusions

Combining evidence from experimental and observational studies derived a significant protective health benefit of dysphagia screening following adult acute stroke for pneumonia, mortality, dependency, and length of stay.

Nonstandard Abbreviations and Acronyms

SLP

speech‐language pathologist

Clinical Perspective

What Is New?

• New high‐level evidence from meta‐analysis data suggests health benefits from early dysphagia screening in adult patients diagnosed with acute stroke.

What Are the Clinical Implications?

• These findings support the use of dysphagia screening protocols applied early after stroke to mitigate unnecessary health consequences of pneumonia, mortality, dependency, and length of stay.

Dysphagia is common following stroke, affecting ≈55% of acute stroke patients,¹ and leads to complications such as aspiration pneumonia,¹ malnutrition,² dependency, and mortality.¹, ³ There has been recent effort to promote early identification of dysphagia with screening as a critical first step to promote improved recovery.

Best practice stroke guidelines⁴, ⁵ state level 2 evidence to support early detection for screening, yet practice remains varied. Despite available psychometrically sound screening tools,⁶ adherence to dysphagia screening across acute stroke institutions in both the United States and Canada has ranged from 56.7%⁷ to 80.8%.⁸ The current evidence, based on only randomized controlled trial (RCT) data, leaves the healthcare professional unclear as to best practice for patients with stroke. We posit that reviews limited to RCT data may have inadvertently overlooked more convincing clinical data.

To address the uncertainty related to dysphagia screening benefit in patients with stroke, we conducted a systematic review of the literature that extends beyond RCT data and also includes high‐quality observational evidence. The specific aim of this study was to assess whether early detection for dysphagia with bedside screening administered to adult patients with stroke by a nondysphagia expert reduces the frequency of health‐related complications such as pneumonia, dependency, mortality, and length of hospital stay (LOS) compared with similar patients with no or relatively less rigorous early detection.

METHODS

The data that support the findings of this study are available from the corresponding author upon request. This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses protocol⁹ for treatment benefit and was not a candidate for institutional ethical review. Our search was directed by the following operational definitions: stroke included any stroke etiology; dysphagia was impaired swallow physiology of the upper aerodigestive tract; and screening was bedside testing by a clinician other than a speech‐language pathology (SLP) dysphagia expert.

Search Methodology

An original search was conducted in 2013. We searched for relevant peer‐reviewed abstracts in the following electronic databases: MEDLINE (1946–January 2013), PsycINFO (1806–January 2013), MEDLINE In‐process (February 2013), EMBASE (1974–January 2013 ), CINAHL (1990–2013 February), Cochrane Database of Systematic Reviews (2005–December 2012), and Cochrane Central Register of Controlled Trials (January 2013). Across databases, the general search strategy included deglutition, deglutition disorders, dysphagia, swallowing, swallowing disorders, oropharyngeal, stroke, poststroke, cerebral vascular, screen, and early detection. This search was reproduced and expanded to include all literature published up to December 2019. All strategies were rerun using the same databases, with the exception of 1. CINAHL was replaced with EMCARE from Ovid, as the CINAHL database changed from the original search. Rather than omitting a database, we added EMCARE, which targets nursing and allied health. Other differences between the original and expanded search included (1) MEDLINE In‐process expanded from original search to include e‐publication ahead of print in reproduced search, and (2) a new subject heading "stroke rehabilitation" was added to the databases MEDLINE and EMBASE. The complete search strategy for MEDLINE for both the original and reproduced searches is detailed in Data S1. A manual search was also conducted in researchers' personal libraries. The yield across all data sources and from both searches were merged and duplicates removed resulting in a final unique citation count to December 2019.

Study Selection

Two independent raters excluded citations if they had no abstract; no human participants; were classified as a tutorial or review; were conference proceedings; had <10 eligible patients; had more than 10% of eligible subjects <18 years of age; had more than 10% of eligible subjects with a diagnosis other than stroke; made no mention of dysphagia screening and/or clinical assessment by a non‐speech‐language pathologist; or made no mention of a health‐related outcome (such as pneumonia, nutrition/hydration, death or dependency, LOS). Discrepancies were resolved by consensus. All other abstracts were accepted and brought to full article review. Additional full article review exclusion criteria included no comparison outcome data between the 2 dysphagia screening groups and systematic reviews. Discrepancies for full article exclusion were resolved by consensus between the 2 raters who were blinded to one another's ratings. The kappa coefficient and associated 95% CI was calculated to determine interrater reliability at both the abstract and full article review phase. Studies included in related systematic reviews were included if they met our inclusion criteria and were unique to our yield.

Data Extraction

Data were extracted descriptively from included studies by 1 reviewer and cross‐checked for accuracy by a second rater: sample size, age, sex, stroke type and severity, study design, control and experimental comparison details, dysphagia screening details, number of failed screenings, health outcome details, number with poor health outcomes by type, and, if reported, health outcome adjusted point estimates with 95% CIs. To identify homogenous subgroupings, studies were stratified by control and experimental comparison details.

Quality Assessment

All accepted studies were assessed for methodological quality by 1 reviewer and cross‐checked by a second rater according to Cochrane's Risk of Bias and included the following domains: selection bias, detection bias, attrition bias, and reporting bias.¹⁰ Discrepancies were resolved by consensus.

Statistical Analysis

Homogeneous data related to the target outcomes were entered into RevMan, version 5.3.¹¹ For experimental groupings with varied screening time, we compared results between the earliest versus latest screen. For the same health outcome, a random effects model was used to derive the pooled estimates for ≥2 studies with similar control and experimental comparisons: first across homogeneous comparisons and then across all comparisons. For reported dichotomous raw outcome data we derived odds ratios (ORs) using the random effects model, whereas for reported continuous raw outcome data we derived the standardized mean differences. At the individual study level, odds ratios were calculated by using the standard equation based on 2×2 tables. All point estimates with 95% CIs were depicted using forest plots. Funnel plots were also produced of effect measures against the inverse of SE to assess for publication bias.

RESULTS

Literature Retrieved

The 2 searches yielded a combined total of 8860 citations. Following duplicate removal, 5996 unique abstracts were screened, of which 494 were accepted for full article review including 7 from related systematic reviews. At the abstract review stage, the kappa coefficient was 0.63 (95% CI, 0.60–0.67) and at the full article review stage, the kappa coefficient was 0.90 (95% CI, 0.77–1.04). Across all full articles screened, 33 articles were deemed eligible and moved forward. The 33 articles were based on 27 studies: 2 articles¹², ¹³ from the HEADPOST (Head Position in Stroke Trial),¹⁴ 3 articles¹⁵, ¹⁶, ¹⁷ from the study by Middleton and colleagues,¹⁷ 2 articles¹⁸, ¹⁹ from the study by Perry and McLaren,¹⁸ 2 articles²⁰, ²¹ by Svendsen and colleagues,²⁰ 2 articles²², ²³ by Odderson and McKenna,²² and 25 were unique studies (see Figure 1 for Preferred Reporting Items for Systematic Reviews and Meta‐Analyses and Data S1 for list of excluded articles).

Figure 1. PRISMA flow diagram.

PRISMA indicates Preferred Reporting Items for Systematic Reviews and Meta‐Analyses and SLP, speech‐language pathologist.

Characteristics of Included Studies

Of the 27 studies, 22 studies³, ¹⁹, ²⁰, ²¹, ²², ²⁴, ²⁵, ²⁶, ²⁷, ²⁸, ²⁹, ³⁰, ³¹, ³², ³³, ³⁴, ³⁵, ³⁶, ³⁷, ³⁸, ³⁹ were observational in design, and 5 studies¹², ¹³, ¹⁵, ¹⁶, ¹⁷, ⁴⁰, ⁴¹ were RCTs. Of the observational studies, 10^* assessed data from prospective stroke registries; 7 studies^† used historical controls to compare screening intervention; 5 studies²⁷, ²⁸, ³⁶, ⁴², ⁴³ assessed data from patient medical records, of which 3 studies²⁷, ⁴², ⁴³ were from single institutions and 2 studies²⁸ from multiple hospital sites. Of all studies, 29.6% (n=8^‡) were conducted in European countries and the sample sizes ranged from 101²⁵ to 143 578³⁴ participants. The combination of acute stroke type varied, with 2 articles from the same study but included different stroke types, thus 8 articles^§ with only ischemic and 18 articles^‖ with mixed ischemic and hemorrhagic strokes. Two studies²⁷, ³² did not specify stroke type. Stroke severity was reported in 21 (77.8%) studies using 1 of 3 measures, namely National Institutes of Health Stroke Scale,⁴⁵ Canadian Neurological Scale,⁴⁶ or Intracerebral Hemorrhage Score.⁴⁷ Eleven studies reported overall median stroke severity scores ranging from mild³, ¹², ²⁶, ³⁴, ³⁵ to moderate.¹⁸, ¹⁹, ³⁰, ³³, ³⁶, ⁴¹ Five comparison groups evolved, which we stratified accordingly as no screening versus screening,^¶ late versus early screening,³, ¹⁶, ²¹, ²⁶, ²⁹ informal versus formal screening,³⁰, ⁴¹ pre‐ versus postscreening,¹⁸, ¹⁹, ²⁵, ³², ³³, ³⁴, ³⁵, ³⁷, ³⁹, ⁴⁰ and pre‐ versus poststroke guideline that included screening¹⁵, ¹⁷, ²², ²³, ²⁷, ⁴², ⁴⁴ (see Table S1).

Results from articles using data from the same study were summarized together, unless the group comparisons were different. In the end, 30 articles reported unique data and were moved to data extraction.

Dysphagia Screening

Across all included articles, the dysphagia screening protocols used for early detection varied, with 14^# declaring no specific tool, 4 articles²⁵, ²⁹, ³², ³⁷, ³⁹, ⁴³ declaring an institution‐specific protocol, 1 of which was composed of 2 bedside water swallowing tests⁴⁸, ⁴⁹ and only 12^** using a previously published tool. Of the articles that used published screening tools: 6 articles¹², ¹³, ¹⁵, ¹⁶, ¹⁷, ³⁶, ⁴⁰ included the Acute Screening of Swallow in Stroke/Transient Ischemic Attack,⁵⁰ 4 articles²⁰, ²¹, ³³, ³⁵, ³⁸ included the Gugging Swallowing Screen,⁵¹ 1 article³⁹ used the Three‐Step Swallowing Screen protocol,⁴⁸, ⁴⁹ and 1 used the MetroHealth Dysphagia Screen.⁵² There was variation in who administered and interpreted the screening tools: 20 articles^†† reported a health professional such as a nurse or doctor, 12 of which^‡‡ clearly specified screening training; 8 articles¹², ¹⁵, ¹⁶, ¹⁷, ²², ²³, ³⁰, ³⁸, ⁴⁰ specified that although SLPs were not the sole screener they were involved in dysphagia screening at various levels, of these 3 articles¹², ¹⁵, ¹⁷, ⁴⁰ used the Acute Screening of Swallow in Stroke/Transient Ischemic Attack⁵⁰ and 2 articles³⁵, ³⁸ the Gugging Swallowing Screen⁵¹; and the screener was unclear in 2 articles.²⁰, ²¹

Five articles compared late versus earlier screening: 3 articles³, ¹⁶, ²¹ used a threshold of 24 hours; 1 article²⁹ <4 hours; and 1 article²⁶ <79 minutes. Two³⁰, ⁴¹ articles compared informal screening without clear instruction versus formal screening with clear instruction. Nine articles compared standard care without screening (pre‐) versus different screening experimental groups (post‐), namely: 2 articles²⁵, ³² provided additional training to screeners and 7 articles¹⁸, ¹⁹, ³³, ³⁴, ³⁵, ³⁷, ³⁹, ⁴⁰ implemented specific screening protocols. Likewise, 6 articles compared standard care without stroke guidelines (pre‐) versus poststroke guidelines that included screening at varying years post implementation: first year,¹⁵, ¹⁷, ²², ²³, ²⁷, ⁴⁴ second year,²², ²³, ⁴⁴ or third year.⁴²

Screening Adherence

Adherence was relevant for the 20 articles^§§ that allocated screening to all patients in the experimental groups. Of these, 12 articles ^‖‖ reported screening adherence, with rates that ranged from 18.2%⁴⁴ to 100.0%.³³

Dysphagia Frequency

The incidence of dysphagia from screening results was reported in only 8 articles,^¶¶ with the frequency ranging from 20.5%²⁵ to 62.3%.¹⁹

Reported Health Outcomes

Across the 30 included articles, the benefit of dysphagia screening was measured with various health‐related outcomes, including pneumonia in 20,^## mortality in 18,*** dependency in 13,††† and LOS in 13‡‡‡ (see Table). Pneumonia was captured within hospital stay in 16 articles§§§ but for the remaining 4 articles²², ²³, ²⁸, ⁴², ⁴³ time was not specified. Eleven articles‖‖‖ identified pneumonia from both clinical and radiographic findings, 3 articles²⁵, ²⁶, ⁴⁴ from only clinical findings and 1 article³⁷ from only radiographic findings. Of the 20 articles that captured pneumonia, only 1 article¹⁹ defined it broadly as infection, whereas all others¶¶¶ were specific to infection secondary to aspiration.

Table 1.

Criteria for Defining Health Outcomes

Study	Pneumonia*	Mortality	Dependency
No screening vs screening
Abilleira et al, 201224	N/A	30 and 90 d	N/A
Gonzalez‐Suarez et al, 201828	NR	N/A	N/A
Joundi et al, 2019† 31	N/A	N/A	Enteral tube feeding dependency within hospital stay
Munoz‐Venturelli et al, 201912	N/A	90 d	mRS≥34 at 90 d
Ouyang et al, 201913	Presence of 3 or more of following features: new or worsening cough, increased respiratory rate, oxygen desaturation, fever >38ºC, leukocytosis or leukopenia, and purulent secretions, rales or bronchial breath sounds with positive radiological abnormalities	N/A	mRS 3–6‡ at 90 d
Phan et al, 201936	Clinical findings of: fever ≥38ºC, consolidation on cheat radiograph and use of antibiotic medications	N/A	N/A
Svendsen et al, 2009§ 20	N/A	N/A	N/A
Rather et al, 201043	NR	N/A	N/A
Teuschl et al, 201838	Presence of clinical symptoms (eg, cough, purulent sputum) and clinical signs (eg, fever, rales, bronchial breath sounds or elevation of inflammatory markers in laboratory tests) confirmed by at least 1 chest radiograph within 7 d after stroke Hospital‐associated pneumonia was defined as pneumonia diagnosed >7 d of admission	In hospital, on stroke unit	N/A
Late screening vs early screening
Al‐Khaled et al, 20163	Based on a combination of clinical presentations, radiologic signs detected on a chest radiograph, and blood test results (C‐reactive protein and leukocytes)	N/A	mRS ≥ 2–5 at discharge and at 90 d
Bray et al, 201726	Stroke associated pneumonia was determined by physician and defined as the administration of antibiotics for a new clinical diagnosis of pneumonia <7 d after admission	30 d	N/A
Han et al, 201829	Pneumonia was diagnosed by clinical examination that was supported and confirmed by biochemical, microbiological, and radiological evidence <7 d of admission	In hospital	mRS score of 3–5 at discharge
Middleton et al, 201916	N/A	N/A	mRS ≥24 at 90 d
Svendsen et al, 2014‖ 21	N/A	N/A	N/A
Informal screening vs formal screening
Hinchey et al, 200530	Either clinical finding of rales or dullness to discussion and 1 of the following: purulent sputum, isolation of organism, or chest radiograph showing evidence of infiltrate/consolidation/cavitation OR pleural effusion and 1 of following: purulent sputum, isolation of agent or antibody evidence of agent	In hospital	N/A
Rai et al, 201641	Aspiration pneumonia was diagnosed if: (1) fever >38ºC, (2) purulent secretions/sputum, (3) pulmonary consolidation on physical examination, (4) presence of a new or progressive radiographic infiltrate, or (5) leukocytosis (>12 000/mm3) or leukopenia (<4000/mm3)	In hospital and 90 d	mRS at discharge and mRS ≤2 at 90 d
Prescreening vs postscreening
Bravata et al, 200925	Pneumonia events were identified based on medical record documentation of pneumonia diagnosis by clinician	In hospital	Functional status at discharge¶
Kampman et al, 201532	N/A	In hospital; within 2 wk of admission	N/A
Lopes et al, 201833	Presence of a pulmonary consolidation or infiltration in thoracic imaging and 3 or more of the following: fever ≥38ºC, productive cough, abnormal respiratory examination, arterial hypoxemia, elevation of blood inflammatory markers, and microbiological identification of a relevant pathogen	In hospital and 90 d	90 d mRS scores, cutoff not specified
Middleton et al, 201940	N/A	N/A	Barthel Index score ≥95 (independence) at 90 d mRS ≥24 at 90 d
Paley et al, 201834	N/A	30 d	N/A
Palli et al, 201735	At least 1 of the following: fever >38ºC, leukopenia, and for adults ≥70 y, altered mental status with no recognized cause. At least 2 of the following: new onset of purulent sputum or increased respiratory secretions or increased suctioning requirements; new or worsening cough or dyspnea or tachypnea; rales, crackles or bronchial breath sounds; worsening gas exchange Positive chest x‐ray examination was a prerequisite	In hospital	N/A
Perry & McLaren 200019	Infectious rates included chest infections (diagnosed clinically by medical teams and treated with antibiotics), aspiration pneumonia or sepsis (treated with antibiotics)	In hospital	Barthel Index score at discharge
Schrock et al, 201837	A new infiltrate on chest radiogram treated with antibiotics	In hospital	N/A
Yeh et al, 201139	Nosocomial pneumonia: (1) rales in breathing sound examination or dullness in chest percussion or (2) radiological evidence of new infiltration, consolidation, cavitation, or pleural effusion, and at least 1 of the following: (1) new onset of purulent sputum, (2) positive blood culture, and (3) positive sputum culture	In hospital	N/A
Preguideline vs postguideline
Burgess et al, 201242	NR	In hospital and 6 mo	Barthel Index score at discharge
Di Matteo et al, 200427	N/A	Death at disposition	N/A
Middleton et al, 201715; 201117	Discharge diagnosis was based on documentation of aspiration pneumonia (coding from International Classification of Diseases, Tenth Revision)	90 d and 3–5 y	mRS ≥24 at 90 d
Odderson et al, 199322; 199523	NR	N/A	N/A
Pinero‐Saez et al, 201844	Pneumonia diagnosis in medical discharge report	N/A	N/A

mRS indicates Modified Rankin Scale; N/A, not applicable; and NR, not reported.

^*Time: within hospital stay unless otherwise specified.

^† Only reported use of enteral feeding.

^‡ Includes combined dependency and mortality scores.

^§ Only reported length of hospital stay.

^‖ Only reported length of hospital stay.

^¶ Dichotomous outcome: functional=independent with all activities of daily living; not functional=not independent with all activities of daily living.

Likewise, mortality was captured at varying time points: captured anytime within hospital stay in 13 articles###; 30 days poststroke in 3 articles²⁴, ²⁶, ³⁴; 90 days poststroke in 5 articles¹², ¹⁷, ²⁴, ³³, ⁴¹; 6 months poststroke in 1 article⁴²; and, 3 to 5 years poststroke admission in 1 article.¹⁵ Dependency was captured at discharge in 6 articles³, ¹⁸, ¹⁹, ²⁵, ²⁹, ⁴¹, ⁴² and 90 days follow‐up poststroke admission in 8 articles³, ¹², ¹³, ¹⁶, ¹⁷, ³³, ⁴⁰, ⁴¹; and captured using the Modified Rankin Scale⁵³ in 9 articles,**** with varying impairment score cutoffs, at either 2³, ¹⁶ or 3.¹², ¹³, ²⁹, ⁴¹ Five articles¹², ¹³, ¹⁶, ¹⁷, ⁴⁰ combined mortality and dependency as a composite score based on the Modified Rankin Scale and were stratified as a dependency outcome. The Barthel Index⁵⁴ was used in 3 articles,¹⁹, ⁴⁰, ⁴² 1 of which⁴⁰ defined impairment as a dichotomous score of ≥95 and 2 articles¹⁹, ⁴² reported continuous data. Enteral feeding was captured as a surrogate measure of dependency in 1 article³¹. LOS was captured in 13 articles¹⁷, ¹⁸, ¹⁹, ²⁰, ²¹, ²², ²³, ²⁹, ³³, ³⁵, ³⁹, ⁴¹, ⁴², ⁴³, ⁴⁴ and defined as the duration from stroke hospital admission to discharge. LOS was reported as a continuous outcome in 13 articles,¹⁷, ¹⁸, ¹⁹, ²⁰, ²¹, ²², ²³, ²⁹, ³³, ³⁵, ³⁹, ⁴¹, ⁴², ⁴³, ⁴⁴ with mean and SD scores in 7 articles¹⁷, ¹⁸, ¹⁹, ²¹, ²², ²³, ³⁹, ⁴³, ⁴⁴ and median and range scores in ​5 articles²⁰, ³³, ³⁵, ⁴¹, ⁴². One²⁹ article reported LOS as a dichotomous outcome with a cutoff score of 3 weeks.

Methodological Quality

Randomized Controlled Trials

The risk for selection bias was low across all RCTs but 1 article⁴¹ that based assignment of patients to study arms on hospital ward at admission. All RCTs had high risk of detection bias as raters of dysphagia screening were not blinded to stroke severity¹², ¹³, ¹⁵, ¹⁶, ¹⁷, ⁴⁰, ⁴¹ and/or health outcomes.¹³, ¹⁵, ¹⁶, ¹⁷, ⁴¹ Risk for attrition bias was low for RCTs, with only 2 RCTs¹⁵, ¹⁶, ¹⁷ at high risk owing to low accountability of all participants. All RCTs presented with high risk for reporting bias for dysphagia screening as they used tools previously reported to have insufficient reliability and validity.⁶ Only 1 trial⁴⁰ provided sufficient detail enabling replication of the screening protocol. Three trials¹², ¹³, ⁴¹ did not report screening adherence. In contrast, the risk of reporting bias across all 6 RCTs was low with regard to health‐related outcomes. All trials either operationally defined their outcomes or used measures that were previously proven reliable and accurate. Outcome data were not complete in 2 trials¹⁶, ¹⁷ and in 1 trial¹² it was unclear whether outcomes were identified using the same surveillance schedule across all patients.

Observational articles

The risk for selection bias was low as all observational articles specified inclusion criteria, and all but 6 clearly specified consecutive enrolment. There were 5 articles²⁵, ³⁰, ³², ³⁸, ⁴⁴ articles that did not report data adjusted for covariates and thus presented with high risk of bias for differences between control and experimental groups. All observational articles had high risk of detection bias as no study reported rater blinding for dysphagia screening or health‐related outcomes. Similar to RCTs, observational articles had low risk of attrition bias, with only 1 study²⁴ presenting with high risk of bias within this domain. All observational articles presented with high risk of reporting bias owing to poorly detailed screening protocols,†††† use of screening tools that have not been shown reliable or valid,²¹, ²⁷, ³³, ³⁵, ³⁸, ³⁹, ⁴², ⁴³ and/or inconsistent screening protocols across study patients.²⁶, ²⁷, ³¹, ³⁶, ⁴² The risk of reporting bias for health‐related outcomes was low. Across all articles, only 2 of them³⁵, ³⁷ were unclear whether surveillance schedules for health outcomes were similar for all patients, and only 1 article¹⁹ was unclear on accountability of health outcomes (see Figure 2).

Figure 2. Critical appraisal of included articles.

Screening Benefit to Health and Related Outcomes

The reported raw data and adjusted estimates between control and experimental groups for each comparison grouping is depicted descriptively (see Table S1). Of the 20 articles that reported pneumonia, 8 articles³, ¹³, ²⁶, ²⁸, ²⁹, ³³, ³⁹, ⁴¹ reported adjusted estimates: 6 articles³, ²⁶, ²⁸, ²⁹, ³⁹, ⁴¹ favoring experimental groups and 2 articles¹³, ³³ favoring control groups. Of all 18 articles that reported mortality, 7 reported adjusted estimates: 4 articles¹², ¹⁵, ²⁹, ⁴¹ favoring experimental group. Of all 13 articles that reported dependency, 5 articles³, ¹², ¹³, ¹⁶, ²⁹ reported adjusted estimates, of which 3 articles¹², ¹³, ¹⁶ reported a composite score including dependency and mortality. All 5 favored the experimental group. Lastly, of all 13 articles that reported LOS, 2 articles²⁰, ²⁹ reported adjusted estimates for LOS, both favoring the experimental group.

Health‐related outcomes were reported as raw data in 21 articles and were therefore pooled by outcome domain and comparison grouping for all articles (Figure 3A through 3D), and then for only the subgroup of articles that clearly specified screening was administered by a non‐SLP dysphagia expert (Figure 4A through 4D). A random effects model was used based on the assumption that each sample came from a different population with different screening comparison groups and that the effects in the different populations may also differ. This analysis may have led to more conservative results.⁵⁵, ⁵⁶

Figure 3. Meta‐analysis of health outcomes.

A, Pneumonia; (B) mortality; (C) dependency; (D) length of hospital stay.

Figure 4. Meta‐analysis of health outcomes, screening by non‐SLP only.

A, Pneumonia, screening by non‐SLP only; (B) mortality, screening by non‐SLP only; (C) dependency, screening by non‐SLP only; (D) length of hospital stay, screening by non‐SLP only. SLP indicates speech‐language pathologist.

Pneumonia

The overall pooled OR for pneumonia events was 0.57 (95% CI, 0.45–0.72), favoring the experimental group (see Figure 3A). However, within comparison grouping, results were mixed, with pooled ORs favoring the experimental group in only 3: late versus early screening (OR, 0.39; 95% CI, 0.24–0.61), informal versus formal screening (OR, 0.43; 95% CI, 0.27, 0.69), and pre‐ versus postscreening (OR, 0.58; 95% CI, 0.37–0.92). Likewise, the overall pooled effect for pneumonia in articles that clearly excluded SLPs as screeners also favored the experimental group with an overall OR of 0.56 (95% CI, 0.43–0.74) (see Figure 4A), with the pooled OR favoring the experimental group in only the late versus early screening subgroup (OR, 0.39; 95% CI, 0.24–0.61).

Mortality

The overall pooled OR for mortality was 0.52 (95% CI, 0.35–0.77), favoring the experimental group (see Figure 3B). However, similar to pneumonia, results within comparison groupings were mixed with only 3 favoring the experimental group: no screening versus screening (OR, 0.29; 95% CI, 0.09–0.95), late versus early screening (OR, 0.22; 95% CI, 0.20–0.23), and informal versus formal screening (OR, 0.55; 95% CI, 0.37–0.82). The overall pooled effect for mortality in articles that excluded SLPs as screeners did not significantly favor the experimental group although the late versus early screening comparison grouping did (OR, 0.22; 95% CI, 0.20–0.23) (see Figure 4B).

Dependency

The overall pooled OR for dependency was 0.54 (95% CI, 0.35, 0.85), favoring the experimental group (see Figure 3C). Results within comparison groupings were mixed with only 3 favoring the experimental group including no screening versus screening (OR, 0.72; 95% CI, 0.60–0.88), late versus early screening (OR, 0.16; 95% CI, 0.05–0.58) and pre‐ versus postguideline (OR, 0.56;, 95% CI, 0.44–0.71). Similarly, the overall pooled effect for articles that excluded SLPs as screeners favored the experimental groups (OR, 0.61; 95% CI, 0.40–0.92) with derived pooled estimates for 2 comparison groupings favoring the experimental group: no screening versus screening (OR, 0.65; 95% CI, 0.57–0.75), and late versus early screening (OR, 0.31; 95% CI, 0.22–0.44) (see Figure 4C).

Length of Hospital Stay

The overall pooled estimates for LOS favored the experimental group with a standardized mean difference of −0.62, (95% CI, −1.05 to −0.20). Results within comparison groups were mixed with only early versus late screening favoring the experimental group (standardized mean difference = −0.28; 95% CI, −0.33 to −0.23). In contrast, the overall pooled effect or derived pooled estimates for comparison groupings for articles that excluded SLPs as screeners were not significant.

See Figure 5A through 5D for funnel plots. There was no obvious publication bias detected as represented visually by the symmetrical funnel plots.

Figure 5. Funnel plots for health outcomes.

A, Pneumonia; (B) mortality; (C) dependence; (D) length of hospital stay. OR indicates odds ratio; and SMD, standardized mean difference.

DISCUSSION

The primary goal of this review was to systematically evaluate available evidence assessing the benefit of dysphagia screening on health outcomes in adults poststroke. Our findings yielded 30 unique articles, including 6 RCTs and 24 observational studies with 5 different comparison groupings. Despite the heterogeneity among study design and experimental groupings, and the risks for bias at the article level, our pooled estimates suggest that early dysphagia screening of adult patients admitted with acute stroke reduces their risk for pneumonia, mortality, overall dependency, and LOS.

Through meta‐analysis, we identified pooled estimates favoring the experimental group for pneumonia, mortality, overall dependency, and LOS, which persisted for pneumonia and dependency within the subgroup of articles that included only non‐SLP screeners. This finding provides support to practice guidelines⁴, ⁵ that advocate for screeners to receive training on tool administration and interpretation. Stroke guidelines recommend screening protocols that are reliable and valid, thereby serving to accurately identify patients at risk for dysphagia and prioritize their referral for a comprehensive dysphagia assessment by an SLP. Interestingly, of all the 30 retrieved articles only 12 used screening protocols with published protocols, none endorsed for sufficient reliably and validity.⁶ Despite these limitations, our findings identified screening benefit to health outcomes poststroke when screening was offered soon after admission and by a trained screener.

Our findings of dysphagia screening benefit to health align with the work from others,⁵⁷ including 1 article⁵⁸ from our group published over 20 years ago. Our findings however contradict more recent suggestion of insufficient evidence.⁵⁹ Specifically, we identified 3 recently published trials¹², ¹³, ⁴⁰ that showed benefit from screening on mortality and dependency, although not pneumonia. Furthermore, our inclusion of observational articles uniquely represented large stroke data sets and several multi‐institutional findings. From these clinical studies we identified select features of dysphagia screening that benefit the health of patients with stroke, especially highlighting positive impact from earlier versus later screening. By virtue of its broader scope, our meta‐analysis can perhaps be considered the first real‐world assessment for the benefit of dysphagia screening in patients with stroke.

Despite our effort to maintain a rigorous literature review, there are methodological limitations in the available literature and within our methodology that are important to acknowledge. The high risk for detection and reporting bias within RCTs presents concern for study duplication. The bias most critical within observational articles was likely selection bias, in that the cohort comparison groupings may have offered screening to patients with more severe strokes types thereby disadvantaging the screened experimental group to more severe dysphagia and subsequently poorer health outcomes.³¹, ³⁶, ³⁸ This potential bias might explain why some patients who were screened were at increased odds for dependency on enteral feeding.³¹ Historical controls¹⁸, ¹⁹, ²², ²³, ³⁵, ³⁷, ⁴⁴ captured data at 2 different time points presenting with high risk for bias related to differing potential confounding factors. Undoubtedly, these methodological limitations are serious considerations at the single article level, but perhaps by pooling their data across homogenous subgroupings of similar screening comparison groups, our analysis offers the first pragmatic evidence of screening benefit. In addition, there were some articles where it was unclear whether screening was by trained health professionals other than SLPs. It could be argued that a dysphagia screen by an SLP is not truly a screen but rather an assessment, so we thought it was important to stratify data by this feature. Interestingly, a clear benefit to health for both pneumonia and dependency persisted in the subgroup that clearly used trained nondysphagia experts.

Conclusions

Altogether, this meta‐analysis provides novel findings that support the health outcome benefits of dysphagia screening in adults poststroke. This benefit was especially realized when screening was administered by trained screeners soon after admission to hospital. Critical next steps include comparing and contrasting existing psychometrically sound screening protocols to identify those that provide the most benefit to the health of patients with stroke. The findings from this study establish the necessary foundation for such a future clinical effectiveness trial.

Sources of Funding

V. Sherman received support from the SickKids Clinician Scientist Training Program. R. Martino received support from Canada Research Chair (Tier II) in Swallowing Disorders.

Disclosures

R. Martino is the lead investigator and copyright holder of the dysphagia screening tool the Toronto Bedside Swallowing Screening Test (TOR‐BSST)^©. The remaining authors have no disclosures to report.

Supporting information

Data S1

Table S1

References 3, 12, 13, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 50, 51, 53, 54, 55

(J Am Heart Assoc. 2021;10:e06051. DOI: 10.1161/JAHA.120.018753.)