Predictors of Percutaneous Endoscopic Gastrostomy Tube Placement in Patients with Severe Dysphagia from an Acute-Subacute Hemispheric Infarction

Abstract

This study investigated the influence of age, National Institutes of Health Stroke Scale (NIHSS) score, time from stroke onset, infarct location and volume in predicting placement of a percutaneous endoscopic gastrostomy (PEG) tube in patients with severe dysphagia from an acute-subacute hemispheric infarction. We performed a retrospective analysis of a hospital-based patient cohort to analyze the effect of the aforementioned variables on the decision of whether or not to place a PEG tube. Consecutive patients were identified using International Classification of Diseases, Ninth Revision (ICD-9) codes for acute ischemic stroke, Current Procedural Terminology (CPT)-4 codes for a formal swallowing evaluation by a speech pathologist, and procedure codes for PEG placement over a 5-year period from existing medical records at our institution. Only patients with severe dysphagia were enrolled. A total of 77 patients met inclusion criteria; 20 of them underwent PEG placement. The relationship between age (dichotomized; < and ≥75 years), time from stroke onset (days), NIHSS score, acute infarct lesion volume (dichotomized; < and ≥100 cc), and infarct location (ie, insula, anterior insula, periventricular white matter, inferior frontal gyrus, motor cortex, or bilateral hemispheres) with PEG tube placement were analyzed using logistic regression analysis. In univariate analysis, NIHSS score (P =.005), lesion volume (P =.022), and presence of bihemispheric infarction (P =.005) were found to be the main predictors of interest. After multivariate adjustment, only NIHSS score (odds ratio [OR], 1.15; 90% confidence interval [CI], 1.02–1.29; P = .04) and presence of bihemispheric infarcts (OR, 4.67; 90% CI, 1.58–13.75; P =.018) remained significant. Our data indicates that baseline NIHSS score and the presence of bihemispheric infarcts predict PEG placement during hospitalization from an acute-subacute hemispheric infarction in patients with severe dysphagia. These results require further validation in future studies.

The cerebral cortex plays an important role in regulating both the automatic and the volitional phases of swallowing.^1–3 Lesions that affect the cerebral hemispheres, especially strokes, are known to produce dysphagia.^4–7 Dysphagia afflicts many stroke patients, contributing significantly to their overall morbidity and mortality.^7,8 In many such patients, swallowing function improves spontaneously over time, but in some, the disability is more long-lasting.^5,7 The temporal profile of swallowing recovery in patients with hemispheric infarcts remains poorly delineated, however, and variables that influence the recovery of swallowing function have not been well established. This poses a significant management dilemma when caring for these individuals and is a source of unexplained variability in dysphagia outcome studies involving such populations.

Stroke patients with severe dysphagia typically have their oral intake curtailed to prevent aspiration and are fed via a nasogastric tube (NGT) until swallowing function recovers. Although easy to place, an NGT carries risks of tube displacement that may itself promote aspiration, it may lead to infections such as sinusitis or mechanical irritation of the nasopharynx, and can be uncomfortable for the patient. Percutaneous endoscopic gastrostomy (PEG) with feeding tube placement is used in those individuals whose swallowing function does not improve substantially within a few days after stroke. Although PEG is a more desirable method of sustaining patient nutrition over longer periods, it too carries risks of procedural complications, including bowel perforation. Results from a recently concluded large multicenter FOOD trial discourage the early use of PEG placement in dysphagic stroke patients.⁹ Delaying PEG placement allows time to assess a patient’s ability to recover swallowing function. It is a common practice for such patients to be fed via NGT and be monitored in the hospital until a decision is made regarding PEG tube placement. However, a significant number of patients may fail to regain sufficient swallowing function and may eventually require PEG. Early identification of such patients would be beneficial for patients and their caregivers and would also reduce the length of hospital stays and, consequently, the costs of care.

To identify factors influencing the decision for or against PEG placement in hospitalized stroke patients, we need to consider the variables that influence swallowing recovery in the acute-subacute stroke phase. Most previous studies on this topic have investigated associations of specific clinical and swallowing parameters with swallowing recovery.^10–13 In some of these studies, such clinical associations were not adjusted for other important clinical and neuroanatomical variables that might influence swallowing recovery. Several brain regions (most notably the motor cortex, inferior frontal gyrus, insula, anterior insula, and periventricular white matter) have been shown to play a role in regulating swallowing function and/or have been associated with the development of dysphagia after stroke.^1,2,14 These regions’ roles in swallowing recovery are not well understood, however. Mapping of changes in cortical representation of swallowing after hemispheric infarction has demonstrated that swallowing recovery depends critically on reorganization of the contralesional swallowing cortex.¹⁵ Previous studies of functional recovery after stroke found that baseline National Institutes of Health Stroke Scale (NIHSS) score,¹⁶ age,¹⁷ and stroke lesion volume^18,19 have a significant influence on recovery. The impact of these variables on cortical reorganization or recovery of swallowing functions, however has not yet been systematically explored.

The present study was conducted to investigate the impact of important clinical and neuroanatomical variables that predict failure of swallowing recovery and subsequent PEG placement during the acute-subacute phase of stroke recovery. For the purpose of this study, we focused our investigation on a cohort of severely dysphagic stroke patients with hemispheric infarcts, because these stroke survivors would be the most likely to be considered for PEG placement. We analyzed the effect of age, time after stroke onset, clinical stroke severity as assessed by NIHSS score, and neuroanatomical correlates, including lesion location and volume, in predicting PEG placement in this group of hospitalized stroke patients.

Patients and Methods

All patients were identified retrospectively using International Classification of Diseases, Ninth Revision (ICD-9) codes for stroke (433, 43311, 43321, 43331, 43411), Current Procedure Terminology (CPT)-4 codes for swallowing assessment by a speech pathologist (92610, 92611, 74230) and procedure codes for PEG placement (430, 431,432) over a 5-year period from April 1, 2003, to March 31, 2008, from the inpatient population at our hospital. Ischemic stroke patients who underwent formal swallowing evaluation were identified by combining the ICD-9 codes for stroke and swallowing evaluation; this yielded 804 patients. Similarly, a list of ischemic stroke patients who underwent swallowing evaluation and subsequent PEG placement was obtained by combining patients with ICD-9 codes for ischemic stroke, CPT-4 codes for swallowing assessment, and procedure codes for PEG placement; this yielded 66 patients.

We initially screened all patients for the presence of dysphagia and the diagnostic accuracy of a hemispheric infarction. We excluded patients with brainstem infarction, because the aim of this study was to assess predictors of swallowing recovery only in patients with hemispheric infarctions. Among these patients, we excluded those who had other confounding conditions that might have independently produced swallowing impairment, such as Parkinson’s disease, brain tumor, advanced dementia, multiple sclerosis, inflammatory myopathies, myasthenia gravis, and muscular dystrophies. The medical records of the remaining patients were carefully scrutinized for completeness of medical information. Clinical information, including age and details of swallowing assessment, including time delays in evaluation from stroke onset (in days) and NIHSS score, was recorded. In patients lacking a documented specific time of onset, the time last known to be normal was considered the time of stroke onset. In those patients who lacked a recorded NIHSS score in the medical records, an NIHSS score was constructed retrospectively using validated methods.²⁰ Patients who did not have recorded NIHSS score and also lacked sufficient information to enable retrospective construction of NIHSS scores were excluded from the analysis. Similarly, patients lacking adequate information about their swallowing assessment and those with poor-quality brain imaging studies were excluded. This left us with a total of 146 patients with dysphagia after hemispheric infarction as determined by a formal swallowing evaluation. Among these 146 patients, we identified patients with severely impaired swallowing who might have been considered for PEG placement. We defined severe dysphagia as the complete absence of oral intake in accordance with recommendation of the speech pathologist and/or the presence of significant aspiration on swallowing tasks. Seventy-seven patients had severe dysphagia after an acute hemispheric infarction, had complete clinical and imaging data for the purposes of this study, and had no other confounding conditions for dysphagia. This group of 77 patients served as the final sample for study analysis. We dichotomized this group into a PEG group (20 patients) and a no-PEG group (57 patients). Two patients in whom PEG placement had been recommended but had been declined were included in the PEG group.

Clinical data, including age and NIHSS score, were collected. Imaging studies, including brain magnetic resonance imaging (MRI) and computed tomography (CT) scans of the head, were reviewed by investigators (S.K., A.A., and R.G.) who were blinded to the clinical data. We used the customized Web-based software package Image J (developed by Wayne Rasband, National Institutes of Health; available at http://rsb.info.nih.gov/ij) to calculate lesion volume. Lesion location was recorded in binary form (present/absent) through comparison with a standardized brain MRI atlas (available at http://www.med.harvard.edu/AANLIB/home.html). We specifically assessed for the presence of lesions in brain regions that have been previously implicated in regulating swallowing function, including the motor cortex, inferior frontal gyrus, insula, anterior insula, subcortical periventricular white matter.^1,2,14 We also assessed for the presence of bihemispheric lesions (new or old). All regions were scored separately in each individual patient. Lesion volume and location were analyzed on diffusion-weighted imaging (DWI) sequences of brain MRI images obtained between 24 and 72 hours after symptom onset (n = 69) and plain head CT scans performed between 48 and 72 hours after symptom onset (n = 8).

Variables were analyzed by univariate and then multivariate logistic regression using SAS software (SAS Institute, Cary, NC). For the purpose of statistical analysis, 2 continuous variables—age and infarct volume—were dichotomized into 2 groups each (age < and ≥75 years; infarct volume < and ≥100 cc) to improve clinical interpretability. NIHSS score and time from stroke onset to initial swallowing evaluation were retained as continuous variables. All remaining variables were entered in the analysis as binary variables (present/absent). Multivariate modeling included any variable found to have a P value ≤.10 in univariate analysis; the binary variable periventricular white matter infarcts was included in the final model based on clinical considerations. Our hospital’s Institutional Review Board approved the study design.

Results

The baseline characteristics of our study cohort are tabulated in Table 1. Most of our patients had a partial anterior circulation stroke, were elderly (median age, 76 years) and female, and had sustained a moderately severe stroke with a relatively large infarct volume. The time from stroke onset to swallowing evaluation is presented in Table 2. Three patients in the PEG group and 4 patients in the no-PEG group had an abbreviated initial bedside swallowing evaluation because of drowsiness; only 1 patient in the PEG group had a limited final swallowing evaluation, due to persistent lethargy. Patients who eventually underwent PEG placement usually had more than 2 swallowing evaluations, whereas those who did not typically had 2 evaluations. The important characteristics of dysphagia seen on initial evaluation in our sample are listed in Table 2. Many patients had more than 2 major impairments, with aspiration and delays in the oral and pharyngeal phases of swallowing the most common abnormalities in both groups.

Table 1.

Baseline characteristics of the study cohort

Characteristic
Median age, years	76
Female sex, n (%)	50 (64.9)
Median NIHSS score	8
Mean stroke volume, cm3	43.02
Oxfordshire Stroke Classification, n (%)
Total anterior circulation stroke	2 (2.6)
Partial anterior circulation stroke	65 (84.4)
Posterior circulation stroke	6 (7.8)
Lacunar stroke	4 (5.2)
Stroke subtype based on TOAST criteria, n (%)
Large vessel atherosclerosis	15 (19.5)
Cardioembolism	34 (44.2)
Small vessel occlusion	4 (5.2)
Stroke of other determined etiology	3 (3.9)
Stroke of undetermined etiology	21 (27.3)

Table 2.

Important time parameters and dysphagia characteristics seen on swallowing evaluations in the PEG and no-PEG groups

	PEG (n = 20)	No-PEG (n = 57)
Median time from stroke onset to first swallow evaluation, days (range)	3 (1–7)	2 (1–9)
Median time from stroke onset to final swallow evaluation, days (range)	7 (4–18)	5 (2–18)
Median time from stroke onset to PEG placement, days (range)	11 (7–19)	NA
Characteristics of swallowing impairment, n (%)
Aspiration	15 (75%)	39 (68%)
Oral/pharyngeal delay	17 (85%)	49 (86%)
Pocketing	4 (20%)	9 (16%)
Anterior spillage	3 (15%)	7 (12%)

Abbreviation: NA, not applicable.

The results of the univariate analysis are summarized in Table 3. In univariate analysis, NIHSS score (P = .005), dichotomized stroke volume (P = .022), and presence of bihemispheric lesions (P =.0055) were significantly associated with PEG placement, whereas age and infarct location involving the inferior frontal gyrus and anterior insula showed trends toward significance. Table 3 presents the variables that qualified to be entered into the multivariate model. In a previous anatomical analysis, the presence of periventricular white matter (PVWM) infarcts was significantly associated with aspiration;²¹ based on this consideration, PVWM was entered into the final model as a binary variable. Because the present study is a preliminary analysis of a small sample of patients, we constructed 90% confidence intervals (CIs) for the odds ratios (ORs) obtained.

Table 3.

Univariate analysis of all predictors for PEG placement in dysphagic stroke patients

Variable	PEG	No-PEG	Maximum likelihood estimate (SE)	P value (univariate χ2 test)
Total patients	20	57
Age, years, median*	80	75	0.0458 (0.024)	.0621
NIHSS score, median	12	8	0.141 (0.05)	.0053
Lesion volume, cc, mean†	75.45	31.65	1.791 (0.786)	.022
Median time from stroke onset to evaluation, days	3	2	1.65 (2.05)	.61
Bihemispheric, n (%)	12 (60)	14 (24.5)	1.5276 (0.55)	.0055
Insula, n (%)	13 (65)	29 (50.8)	0.583	.278
Anterior insula, n (%)	11 (55)	19 (33.3)	0.893 (0.53)	.091
Motor cortex, n (%)	14 (70)	28 (49.1)	0.882 (0.555)	.112
Inferior frontal gyrus, n (%)	13 (65)	27 (47.3)	0.952 (0.555)	.0863
Periventricular white matter, n (%)	10 (50)	18 (31.5)	0.773 (0.53)	.144

^*Age dichotomized (< 75 and ≥ for analysis)

^†Lesion volume dichotomized (< 100 cc and ≥ 100 cc for analysis).

In the final multivariate logistic regression analysis, higher initial NIHSS score (OR, 1.15; 90% CI, 1.02–1.290; P =.04) and the presence of bihemispheric infarcts (OR, 4.67; 90% CI, 1.59–13.76; P =.019) were significantly associated with PEG placement (Table 4). Dichotomized lesion volume and PVWM infarct location, which appears to have been negatively confounded, showed a trend toward associations with PEG, but this did not attain statistical significance. The overall model was significant (likelihood ratio of χ² = 23.21, with 7 degrees of freedom; P =.0016).

Table 4.

Multivariate analysis of significant predictors of PEG placement in dysphagic stroke patients with hemispheric infarcts

Variable	OR (90% CI)	P value (χ2 test)
Age (<75 years and ≥75 years)	2.954 (0.87–9.99)	.143
NIHSS score	1.151 (1.02–1.294)	.047*
Lesion volume (<100 cc and ≥100 cc)	5.755 (1.074–30.836)	.086
Anterior insula	1.359 (0.344–5.361)	.713
Inferior frontal gyrus	0.995 (0.294–3.365)	.995
Periventricular white matter	3.829 (1.196–12.258)	.057
Bihemispheric lesions	4.6 (1.58–13.756)	.018*

^*Denotes statistical significance.

Discussion

The main results of this study demonstrate that initial stroke severity as assessed by NIHSS score and the presence of bihemispheric infarcts are significantly associated with the decision to place a PEG tube in a patient with severe dysphagia due to acute or subacute hemispheric infarction, after controlling and correcting for the possible confounding effects of age, lesion location and time from stroke onset to initial swallowing evaluation. PEG placement also showed a trend toward higher initial stroke lesion volume and the presence of periventricular ischemic lesions in multivariate analysis, although the size of our sample and method of analysis might have been under-powered to detect more modest, but nonetheless significant relationships.

One of the most compelling observations of our analysis is the effect of bihemispheric lesions on PEG placement and thus on the failure of swallowing recovery in the acute-subacute phase after a hemispheric infarction. The presence of bihemispheric infarcts (either acute or chronic ischemic lesions in the contralateral hemisphere) in a stroke patient with severe dysphagia increased the odds of PEG placement by a factor of 4.6 after adjusting for other possible confounding variables. This is consistent with previous investigations showing that swallowing functions are under bihemispheric control and further validates the critical role of the nonlesioned hemisphere in mediating swallowing recovery via compensatory reorganization of its’ swallowing cortex.^3,6,15 Our analysis included both acute and chronic lesions in contralesional hemispheres (ie, the cerebral hemisphere with a smaller acute ischemic lesion or a chronic infarct compared with the lesional hemisphere that had the greater lesion burden), and we were unable to determine the differential effects of acute versus chronic lesions in the contralesional hemisphere on swallowing recovery in our sample. Furthermore, given the nature of our study sample, we were unable to explore the influence of lesion location and lesion volume in the contralesional hemisphere.

We also investigated the influence of specific lesion sites on the decision for or against PEG placement. These lesion sites were carefully chosen based on data from the literature on their roles in mediating swallowing functions.^1,2,14 Although lesions at these sites can impair deglutition, our results indicate that they had no significant impact on the recovery potential of swallowing functions; only the presence of acute periventricular infarcts showed a trend toward association with PEG placement. However, our method did not account for differences in lesion size and extent at these locations or how involvement of multiple such sites in an individual patient affected the chance of PEG placement. The use of newer, more sophisticated imaging techniques, such as voxel-based symptom lesion mapping and diffusion tractography, might help answer these questions in the future.

Total ischemic stroke volume has been shown to be an important predictor of stroke recovery.¹⁸ A recent study found acute stroke volume as assessed by DWI sequences to be a significant variable in predicting outcome in multivariate analysis.¹⁹ In hemispheric stroke patients with dysphagia, Daniels et al²¹ found no significant association between severity of dysphagia and size of the stroke lesion using semiquantitative analysis with head CT scans. We employed more quantitative methods and measured ischemic lesion volumes on DWI sequences of brain MRI images, a more reliable method of detecting cerebral ischemia, in the majority of our subjects. We further adjusted the impact of lesion volume thus obtained for other possible influences in a multivariate analysis. Our results show that, after controlling for the potential confounding effects of age, time from stroke onset, clinical stroke severity based on NIHSS score, and lesion location, stroke lesion volume showed a trend toward association with PEG placement. We dichotomized ischemic lesion volume to improve clinical interpretability in terms of OR; the categories chosen (<100 cc and ≥100 cc) were based on the distribution of infarct volumes in our sample to ensure adequate numbers of patients in the PEG and no-PEG groups. But dichotomizing this continuous variable also might have compromised our statistical power and resulted in a more conservative estimate of the effect of lesion volume on swallowing recovery. A possible solution to this problem might have been to subcategorize age into more groups (eg, 25-cc categories); however, we were constrained by our sample size.

Previous studies of global measures for functional recovery have found that NIHSS score is an important predictor of recovery after stroke after adjusting for other variables.^16,22 Our results demonstrate that baseline NIHSS score also provides significant prognostic information about swallowing recovery. Every 2-point increase in NIHSS score more than doubled the odds of PEG placement, indicating that larger and more severe strokes are more likely to result in PEG placement in patients with severe dysphagia. Our analysis, however does not clarify whether this relationship is uniform or varies across different intervals of NIHSS. It is possible that analyzing different components of the NIHSS, such as severity of dysarthria, neglect, and alertness, might have enhanced our model’s predictive ability, but our sample size was not sufficiently large to incorporate all these variables in our analysis. On the other hand, our investigation included important neuroanatomical substrates, disruption of which would likely have produced these clinical deficits. In addition, the presence of significant cognitive impairment might have influenced the decision regarding PEG placement in some of our patients. Our study does not fully address this question, because our patients did not undergo formal neuropsychological evaluation after stroke. Most patients appeared to have an impairment in the oropharyngeal phase, with significant delays and aspiration, although some exhibited pocketing or impulsivity, which might reflect a poststroke cognitive deficit. Regardless, our findings indicate that higher initial NIHSS score, which correlates with cognitive impairment and social functioning after stroke,²³ is significantly associated with PEG placement in patients with severe dysphagia after a hemispheric infarct.

Several groups have studied the effect of age on swallowing. Overall deglutition slows with aging, and the elderly have delayed initiation of the laryngeal and pharyngeal events, leading to prolonged food bolus transport time.^24–26 These changes in swallowing physiology likely make elderly individuals more susceptible to swallowing impairments and also impair their ability to regain their lost function. Other associated factors, including poor dentition, cognitive impairment, and physical frailty, can contribute as well. In previous stroke recovery studies, age seemed to play an important role in predicting functional recovery, implying a significant influence of aging on the stroke-injured brain’s capacity for compensatory reorganization. In our analysis, we found a trend toward an effect of age on PEG placement in our primary analysis, although this was attenuated once all other variables were included in the model. Similar to the situation with infarct volumes, our method of analyzing this variable might have led us to underestimate its effect in the final model.

As a retrospective analysis of a hospitalized cohort, the present study has some inherent biases and limitations. We have attempted to minimize the impact of these influences by using a systematic method for identifying consecutive ischemic stroke patients with severe dysphagia using ICD-9 and CPT-4 codes; however, it is possible that we were unable to systematically capture all such patients using our approach, which might have inadvertently introduced a selection bias. The other possible source of variability is the methods used to assess the severity of dysphagia and swallowing recovery. The patients in our sample did not undergo a standardized swallowing evaluation using validated clinical swallowing scales, but were identified as having severe dysphagia using systematic clinical swallowing evaluations according to hospital-based protocols. Similarly, recovery of swallowing function was not assessed by repeat standardized swallowing evaluation, although all such patients underwent a repeat swallowing evaluation by the speech pathologist. The use of our endpoint of PEG placement provides a very clinically relevant outcome for assessing swallowing recovery, but makes our analysis less likely to be sensitive for detecting more modest changes in swallowing functions. In addition, the care-givers of these patients might have had varying thresholds for implementing PEG placement, which may have introduced some variability in our results.

Despite these limitations, this multivariate analysis improves our understanding of the influence of important clinical and radiological variables in predicting PEG placement and swallowing recovery during hospitalization from stroke. Identifying these variables is an important step in building predictive models that can capture individual information about a patient and aid in making more educated clinical decisions. In addition, prediction models can be useful in clinical trial design for stratification or severity-adjusted analysis. Our current model exhibits a modest but significant ability to predict PEG placement in severely dysphagic stroke patients with hemispheric infarction. We hope that our findings will spawn larger, prospective studies that are better powered to detect influences of lesion location, lesion size, age, and their interplay.