Validation of an aspiration risk prediction model for Parkinson’s disease based on nomogram: a single-center study

Introduction

Parkinson’s disease (PD), a neurodegenerative disorder, has attracted increasing attention. It is primarily characterized by the progressive loss of dopaminergic neurons, resulting in both motor and non-motor symptoms (Mhyre et al., 2012). Patients with PD often experience a range of non-motor symptoms such as depression, anxiety, and sleep disturbances, which significantly impair their quality of life (Lee & Koh, 2015). Dysphagia is a common clinical symptom in PD, with reports indicating that over 80% of patients develop this condition during the disease course, and nearly all experience it in advanced stages (Frank et al., 2021; Simons et al., 2014; Suttrup & Warnecke, 2016a). Dysphagia markedly increases the risk of aspiration, leading to aspiration pneumonia, which accounts for approximately 70% of deaths in patients with PD (Frank et al., 2021). Therefore, early identification of aspiration risk is crucial for preventing complications and improving patient outcomes.

Clinically, fiberoptic endoscopic evaluation of swallowing (FEES) is considered the gold standard for assessing swallowing function. Previous studies have compared subjective perceptions of swallowing in patients with PD with objective assessments obtained from FEES or video fluoroscopic swallowing studies (VFSS). These studies demonstrated that over 50% of patients with PD who subjectively reported no swallowing disorders exhibited swallowing impairments on objective testing, with up to 15% experiencing silent aspiration (Suttrup & Warnecke, 2016b). However, VFSS has several limitations: it requires specialized equipment, requires the transfer of patients to the radiology department for examination, demands accurate assessment by trained professionals, and exposes patients to radiation. Consequently, the widespread application of VFSS is limited.

Currently, simple methods commonly used in clinical practice to assess the risk of aspiration include the Swallowing Disturbance Questionnaire and the Munich Dysphagia Test-Parkinson’s Disease (MDT-PD), which serve as initial screening tools for swallowing disorders in patients with PD. These instruments have reported sensitivities of 80.5% and 81.3%, and specificities of 82% and 71%, respectively (Manor et al., 2007; Simons et al., 2014). However, they are primarily administered by clinicians and are not integrated into nursing assessment protocols. Moreover, when dysphagia symptoms are not prominent, medical and nursing staff may overlook them. As a result, early assessments are not effectively implemented, and aspiration may only be diagnosed after severe complications have arisen, delaying optimal intervention timing.

Although previous studies have explored aspiration risk in patients with PD, current assessment methods still predominantly rely on clinical experience and basic screening tools (Troche et al., 2016). However, these methods lack standardization and precision, which significantly limits their effectiveness in clinical practice. The existing literature suggests that specific phenotypic features—such as a history of choking and coughing, results of the water-swallowing test, and abnormal facial expressions—are closely associated with an increased risk of aspiration (El Halabi, Arwani & Parkman, 2023; Fasseeh et al., 2021; Mari et al., 1997; Stewart et al., 2024). To date, however, no studies have systematically integrated these factors to create a comprehensive predictive model. To address this need, our study aimed to develop and validate a predictive model for aspiration risk in patients with PD, facilitating early identification and intervention to ultimately enhance their quality of life.

Materials and Methods

Results

Results of predictor selection and model construction

In the training set, univariate analysis initially identified 11 potential predictors from the 33 candidate variables. Subsequent LASSO logistic regression yielded four final predictors: ChokCghHist (choking cough history, OR = 11.427, 95% CI [2.187–59.709]; P = 0.004), WaterSwaT (water-swallowing test, OR = 4.262, 95% CI [1.496–12.140]; P = 0.007), FaceExpr (facial expression, OR = 2.929, 95% CI [1.055–8.134]; P = 0.039), and Barthel (Barthel Index, OR = 0.972, 95% CI [0.950–0.995]; P = 0.017). These emerged as independent factors differentiating aspiration status. Because all VIF values were below 5, severe multicollinearity was ruled out (Table 2).

Table 2:

LASSO logistic regression analysis in the training set.

Variable	VIF	β	OR (95% CI)	p value
ChokCghHist	2.152	2.436	11.427 (2.187–59.709)	0.004
FaceExpr	2.192	1.075	2.929 (1.055–8.134)	0.039
WaterSwaT	2.609	1.450	4.262 (1.496–12.140)	0.007
Barthel	1.670	−0.028	0.972 (0.950–0.995)	0.017

DOI: 10.7717/peerj.20443/table-2

Notes:

Abbreviations

ChokCghHist

choking cough history

FaceExpr

facial expression

WaterSwaT

water-swallowing test

VIF

variance inflation factor

A nomogram was constructed based on these four variables to distinguish aspiration status (Fig. 1) by directly using the LASSO logistic regression coefficients. To estimate aspiration risk, the levels of ChokCghHist, WaterSwaT, FaceExpr, and Barthel were evaluated, and individual scores were assigned according to the nomogram’s scale. The sum of these scores yielded a total score from which the probability of aspiration could be inferred.

Model performance

We estimated the probability of three-day aspiration occurrence for each patient as the output of the LASSO logistic regression model and assessed performance using ROC curves. As shown in Fig. 2 and Table 3, the AUC for the training set was 0.891 and 0.882 (95% CI [0.832–0.945]) after optimism adjustment by bootstrapping. The testing AUC was 0.950 (95% CI [0.872–1.000]), highlighting the model’s robust discriminative ability based on the parameters ChokCghHist, WaterSwaT, FaceExpr, and Barthel. The optimal cutoff score, determined using the Youden index, was 0.551. For the training and testing sets, sensitivity (95% CI) was 0.75 (0.58–0.89, adjusted) and 0.80 (0.61–0.95), respectively, while specificity (95% CI) was 0.89 (0.83–0.97, adjusted) and 1.00 (1.00–1.00), respectively. Correspondingly, accuracy (95% CI) reached 0.82 (0.73–0.91, adjusted) for the training cohort and 0.90 (0.80–0.97) for the testing set, with Brier scores (95% CI) of 0.138 (0.101–0.176, adjusted) and 0.091 (0.052–0.138), respectively.

Table 3:

Accuracy of the prediction score of the LASSO logistic regression model for discrimination between aspiration and non-aspiration.

Variable	Training set (adjusted, 95% CI)	Testing set (95% CI)
AUC	0.882 (0.832–0.995)	0.950 (0.872–1.000)
Cutoff probability	0.551	0.551
Sensitivity	0.749 (0.581–0.891)	0.800 (0.611–0.952)
Specificity	0.894 (0.833–0.972)	1.000 (1.000–1.000)
PPV	0.873 (0.803–0.957)	1.000 (1.000–1.000)
NPV	0.785 (0.687–0.887)	0.833 (0.680–0.960)
PLR	2.073 (0–13.865)	Not estimable
NLR	0.278 (0.118–0.446)	0.200 (0.048–0.389)
TP	46	16
FP	6	0
FN	54	20
TN	14	4
Accuracy	0.820 (0.733–0.908)	0.900 (0.800–0.975)

DOI: 10.7717/peerj.20443/table-3

Notes:

Abbreviations

PPV

Positive Predictive Value

NPV

Negative Predictive Value

PLR

Positive Likelihood Ratio

NLR

Negative Likelihood Ratio

TP

True Positive

FP

False Positive

FN

False Negative

TN

True Negative

The calibration curve (Fig. 3) showed that the predicted probabilities generated by the LASSO logistic nomogram closely approximated the actual incidence of aspiration, indicating a high level of predictive accuracy. Furthermore, decision curve analysis (Fig. 4) confirmed the clinical utility of the model across both the training and external testing sets. In the training set, when the probability threshold for intervention ranged from 20% to 90%, the model’s net benefit substantially surpassed the conventional extremes of treating all patients or withholding treatment entirely (Fig. 4A). These findings were corroborated by the external validation cohort, which exhibited a similarly elevated net benefit across a comparable range of intervention thresholds (Fig. 4B).

For example, at a 20% probability threshold—indicating that clinical intervention is considered once the risk of aspiration reaches this level—the model’s net benefit in the training and testing sets was approximately 0.40 and 0.43, respectively. Both values markedly exceeded the net benefits observed under the treat-all (0.37) or treat-none (0) strategies, highlighting the practical applicability of the model in guiding patient management. Moreover, the model maintained stable and advantageous predictive performance even at higher probability thresholds (70%–90%), a range in which accurate risk stratification is particularly critical in stringent clinical scenarios. These outcomes suggest that this predictive tool not only enhances clinicians’ ability to identify individuals at meaningful risk of aspiration but also provides an evidence-based framework to refine decision-making processes, ultimately supporting more targeted and resource-efficient patient care.

Discussion

PD is commonly characterized by bradykinesia, reduced coordination, and slower movements as its primary clinical features. However, the precise mechanisms underlying swallowing dysfunction in patients with PD remain unclear. Some investigators have suggested that damage to both dopaminergic and non-dopaminergic pathways, which affect the central swallowing center and peripheral neuromuscular elements, contributes to the onset and progression of PD-related swallowing difficulties (Suttrup & Warnecke, 2016b). Impaired motor function in PD can profoundly disrupt the normal swallowing process, leading to reduced laryngeal elevation, prolonged pharyngeal and overall swallowing duration, delayed initiation of the swallowing reflex, and, ultimately, aspiration (Ertekin et al., 2002).

Swallowing difficulties in individuals with PD may develop gradually and can appear at any stage of the disease, regardless of severity. Even in early PD, 95%–100% of patients exhibit swallowing impairments, often in a subclinical form; however, these may go unnoticed by both patients and healthcare providers. Consequently, the association between PD and swallowing dysfunction is frequently overlooked at early stages. By the time patients recognize swallowing problems, many have already progressed to moderate or severe impairment, thereby delaying a formal diagnosis. Notably, there is an apparent discrepancy between patients’ subjective perceptions and objective assessment findings. One study (Palmer et al., 2021) showed that relying solely on patient self-reports yielded a prevalence of approximately 35%, whereas objective evaluation tools yielded a prevalence closer to 85%. Furthermore, a substantial proportion of these patients experience “hidden” swallowing issues (Kalf et al., 2012), and some may even encounter silent aspiration in the early stages of the disease. Collectively, these factors complicate clinical efforts to identify aspiration risk in patients with PD.

Existing research has identified age as an independent risk factor for swallowing dysfunction. Sex, body mass index, disease duration, disease severity, sialorrhea, depression, and cognitive status have also been linked to clinical swallowing problems (Cereda et al., 2014; Ding et al., 2018; Frank et al., 2021; Nienstedt et al., 2019; Simons et al., 2014). Additional studies have suggested that dysarthria, reduced cough reflex sensitivity, and decreased peak cough airflow may be correlated with swallowing impairment and aspiration (Curtis & Troche, 2020; Hegland, Okun & Troche, 2014).

In the present study, the four significant risk factors for aspiration in patients with PD were a history of choking and coughing, a high score on the water-swallowing test, diminished facial expression, and a low Barthel Index. Episodes of choking and coughing indicate prior aspiration events, implying overt swallowing dysfunction or compromised muscle strength and coordination, both of which increase the likelihood of recurrent aspiration as PD progresses. The water-swallowing test is commonly used to evaluate swallowing function by observing whether the patient coughs or chokes during water intake. A score of 3 or higher suggests an underlying swallowing problem that increases the risk of aspiration. Our findings also indicate that reduced facial expression is a risk factor for aspiration. In 1986, Charcot introduced the concept of the “mask face,” proposing facial characteristics as a diagnostic criterion for PD (Goetz, 1986). This “mask face” represents slowed or diminished facial muscular movements, reducing the velocity, elasticity, and coordination of the eyebrows, eyes, cheeks, and lips, thereby affecting overall facial expressiveness. This is a hallmark motor symptom of PD (Bologna et al., 2013; Maremmani et al., 2019).

One proposed mechanism for facial bradykinesia involves widespread pathological changes and neuronal loss within the substantia nigra dopaminergic system, along with insufficient secretion of the inhibitory neurotransmitter dopamine. These alterations result in predominant excitatory neurotransmission, ultimately leading to motor dysfunction and increased muscle tone (Frank et al., 2021; Simons et al., 2014). During early-stage PD, presynaptic dopaminergic function in the putamen, orbitofrontal cortex, and amygdala is altered; dopamine transmission in the mesocortical pathway is impaired; and extensive neural changes throughout the brain contribute to deficits in facial emotion recognition (Argaud et al., 2018). Damage to related telencephalic structures also influences swallowing control. Functional MRI studies have shown that both the bilateral putamen and globus pallidus are activated during swallowing in healthy individuals; however, these regions remain inactive in patients with PD (Suzuki et al., 2003). Additionally, impaired neuromuscular control may lead to muscle rigidity, causing slow movement, stiffness, tremors, and facial motor symptoms (Wang & Tickle-Degnen, 2018).

In our study, the Barthel Index emerged as an independent predictor of aspiration risk, highlighting its significance in evaluating functional independence among patients with PD. The Barthel Index is a widely recognized tool that assesses a patient’s ability to perform essential activities of daily living, such as feeding, dressing, and mobility, with lower scores reflecting greater dependency (Taghizadeh et al., 2020). This functional measure is particularly relevant to aspiration risk, as patients with reduced independence in tasks such as eating or maintaining posture may struggle with swallowing coordination or managing oral secretions—key contributors to aspiration in PD (Gong et al., 2022; Murcia et al., 2010).

The clinical importance of this finding lies in its potential to guide risk stratification and intervention strategies. Our model demonstrated that lower Barthel scores were strongly associated with an elevated likelihood of aspiration, consistent with previous studies linking functional decline to adverse outcomes such as aspiration pneumonia. By incorporating the Barthel Index, our nomogram achieved enhanced predictive accuracy, suggesting that functional status is not merely a covariate but a critical determinant of aspiration risk in this population.

The primary limitation of this study was its small sample size, which may have impacted the generalizability of the findings. Additionally, because the study was conducted at a single medical institution, regional bias may have limited the broader applicability of the developed model and its conclusions. Furthermore, only baseline clinical measurements were considered for model development; dynamic changes in clinical measurements, as well as subsequent interventions such as changes in diet and route of feeding, may contribute to more precise prediction of aspiration. Finally, the study did not consider other potential inhalation risk factors such as oral hygiene, saliva control, anticholinergic burden, and cough-reflex sensitivity, which could further improve the model’s accuracy if included as predictors in future multicenter studies.

Conclusion

In conclusion, this study identified four independent risk factors for aspiration using multivariate logistic regression and developed a nomogram with strong predictive performance. The model exhibited high accuracy and stability across both training and testing datasets, highlighting its substantial clinical value. Future research should aim to expand the sample size, conduct multicenter validation, and incorporate additional clinical indicators to enhance the model’s generalizability and predictive capacity, ultimately providing a more effective tool for clinical risk assessment.

Supplemental Information