An acoustic-only multivariable diagnostic model for Parkinson's disease using bowel sound acoustic signatures Authors: Sicheng Chen, Man Zhang, Surui Fang, Qi Yang, Chuyi Chen, Zhixuan Lei, Yuxia Zhu, Fen Deng, Lijuan Mo, Xi Liu, Changhong Tan, Lifen Chen Published: 2026 ABSTRACT BackgroundEarly detection of Parkinson's disease (PD) remains challenging due to the lack of reliable prodromal biomarkers. Non-invasive markers of gastrointestinal dysfunction are lacking.ObjectivesTo characterize bowel sound acoustic phenotypes in PD and evaluate their potential as non-invasive diagnostic markers by developing an acoustic-based predictive model.MethodsStandardized 5-min bowel sound recordings were collected from 120 PD patients and 100 controls. Thirteen acoustic features were extracted. LASSO (least absolute shrinkage and selection operator) selected predictors for logistic regression models, internally validated with 1000 bootstrap resamples. The acoustic-only model was evaluated without refitting within the development cohort in 48 patients with Hoehn-Yahr stages 1-2 and 100 controls.ResultsPD patients exhibited fragmented, low-energy bowel sound patterns compared with controls. Acoustic features did not differ significantly by constipation status, while two duration measures differed by Hoehn-Yahr stage. A full-feature model incorporating acoustic and biochemical markers achieved an area under the curve (AUC) of 0.8659. The acoustic-only model containing six predictors, three of which were statistically significant, yielded an AUC of 0.7999 (95% confidence interval [CI], 0.737–0.852). Optimism-corrected AUCs were 0.839 and 0.784, respectively. In the early-stage subgroup, the AUC was 0.728 (95% CI, 0.644–0.806), with 68.75% sensitivity and 66.00% specificity at the fixed probability threshold of 0.50.ConclusionsPD patients exhibit distinct bowel sound acoustic phenotypes. Acoustic features provided discrimination between PD patients and controls, including in early-stage disease. This non-invasive profiling approach offers a promising functional biomarker candidate for multimodal PD assessment. Introduction Parkinson's disease is a severe neurodegenerative movement disorder characterized by both motor and non-motor features. Common motor symptoms in Parkinson's patients include bradykinesia, resting tremor, muscular rigidity, and festinating gait. 1 The emergence of motor symptoms is primarily attributed to the loss of dopamine-containing neurons in the substantia nigra pars compacta; however, by this stage, it indicates that a massive loss of substantia nigra dopaminergic neurons may have already occurred and the neural damage has become irreversible. 2 Although current symptomatic treatments can effectively improve motor function, they cannot alter the fundamental progression of the disease. Even a recent clinical trial targeting α-synuclein, such as that for prasinezumab, a monoclonal antibody that binds to aggregated α-synuclein, suggested an improvement in the rate of decline in motor scores, but it failed to meet its primary clinical endpoint. 3 The failure may reflect uncertainty regarding the pathological role of α-synuclein 4 and delayed intervention beyond the optimal therapeutic window, underscoring the importance of early diagnosis in Parkinson's disease. Therefore, the development of reliable and generalizable biomarkers for prodromal or early-stage Parkinson's disease has become a research priority. Non-motor symptoms of PD may precede the onset of motor symptoms and impact patients’ quality of life. Among these, gastrointestinal dysfunction (such as abdominal bloating, nausea, constipation, and delayed gastric emptying) represents one of the most common and earliest manifestations. 5 However, relying solely on these clinical symptoms for diagnosis offers limited sensitivity and specificity, making early and accurate identification challenging. Therefore, searching for biological biomarkers that can objectively reflect early intestinal dysfunction has become a potential avenue for the prodromal diagnosis of PD. The brain-gut axis provides an important rationale for intestinal assessment in PD. In some patients, α-synuclein pathology may originate in the enteric or peripheral autonomic nervous system and spread to the central nervous system, while gut dysbiosis and inflammation may further contribute to neurodegeneration.5,6 According to the “body-first/brain-first” framework, body-first PD is associated with earlier gastrointestinal and autonomic dysfunction, whereas gastrointestinal and peripheral autonomic involvement may occur later in brain-first PD.7,8 Accordingly, the timing and severity of bowel sound abnormalities may differ across PD phenotypes. Nevertheless, assessing intestinal functional status may provide a useful approach for detecting subclinical enteric dysfunction, particularly in patients with a body-first disease pattern. Although the current α-synuclein seed amplification assay (SAA) has demonstrated high predictive value, its reliance on invasive sampling (cerebrospinal fluid or tissue biopsy) and the lack of standardization in testing protocols limit its application in large-scale community screening.9,10 Therefore, developing a non-invasive, simple, and reproducible intestinal function assessment tool holds important practical significance for advancing the prodromal diagnosis of PD. Here, we characterized bowel sound acoustic phenotypes in 120 PD patients and 100 controls. PD patients showed a distinctive high-frequency, low-energy pattern, while no significant difference was detected between PD patients with and without constipation. An acoustic-only multivariable model containing six predictors achieved an AUC of 0.7999, supporting further evaluation of bowel sound analysis as a non-invasive research tool. Patients and methods Study population This study consecutively enrolled PD patients diagnosed according to MDS criteria 11 and controls. All PD patients and controls were inpatients. PD patients were recruited from the Department of Neurology at the Second Affiliated Hospital of Chongqing Medical University between (Sep, 2024) and (Apr, 2026). Controls were admitted to the same department for migraine, tension-type headache, benign paroxysmal positional vertigo, or vestibular neuritis. Bowel sound recordings were obtained after clinical recovery or complete resolution of the presenting symptoms, with no gastrointestinal symptoms or balance impairment. PD patients and controls were matched for age (±3 years) and sex, all aged 50–80 years. Selecting controls from the same neurology department was intended to make the hospital environment, fasting conditions, recording workflow, laboratory testing, and equipment handling as comparable as possible between groups. Exclusion criteria for both groups included severe cardiovascular or respiratory disease, severe cognitive impairment or psychiatric disorders, prior antipsychotic or neurotoxin exposure, severe head trauma, abdominal or gastrointestinal surgery, functional or organic gastrointestinal diseases, systemic conditions affecting intestinal motility (e.g., diabetes, thyroid dysfunction), other neurodegenerative disorders (e.g., MSA, PSP), severe orthostatic hypotension, use of anticoagulants or antiplatelet agents; current use of opioid analgesics, anticholinergics, prokinetics, or laxatives; and pregnancy or lactation. Patients taking antibiotics or antipsychotics during the study period were also excluded. For controls, additional exclusions applied: use of NSAIDs, triptans, antihistamines, or benzodiazepines within 72 h before recording; chronic headache or dizziness (>1 year, ≥15 episodes/month); headache/dizziness with autonomic symptoms; and confirmed vestibular migraine. (Figure S1) Controls were screened through medical history and neurological examination to exclude clinically apparent PD and other neurodegenerative disorders. All controls were free of gastrointestinal symptoms at the time of bowel sound recording. Assessment of potential non-motor features of PD included anosmia/hyposmia and symptoms suggestive of REM sleep behavior disorder. No anosmia or hyposmia was identified through history and clinical examination, and all controls screened negative on the REM Sleep Behavior Disorder Screening Questionnaire (RBDSQ). Olfactory assessment did not include a validated smell test, and polysomnography was not performed. This study was approved by the Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University, and written informed consent was obtained from all enrolled patients and volunteers. Principles of bowel sounds monitoring and analysis Bowel sounds were recorded in a quiet room using an electronic stethoscope (44,100 Hz). Participants in both groups ate lunch between 11:30 and 12:30, and bowel sounds were recorded after a 4–5 h fast following lunch. Levodopa preparations and amantadine were withheld for at least 24 h before recording, whereas dopamine receptor agonists, including pramipexole and ropinirole, and monoamine oxidase B inhibitors, such as rasagiline, were withheld for at least 72 h. These standardized withholding periods were intended to reduce acute medication effects rather than achieve complete pharmacological washout. With subjects supine, the stethoscope was placed 4–5 cm right of the umbilicus; after a 5-min acclimation, sounds were recorded for 5 min. Audio data were transmitted via wired connection to avoid signal degradation. Signal processing and feature extraction were performed using Python 3.13.2 (wave, numpy, pandas, scipy). Detailed protocols are provided in the Supplementary Methods. Recorded bowel sounds were processed using Python 3.13.2 (libraries: wave, numpy, pandas, scipy). Original bowel sound (Figure S2(a)) signal preprocessing employed short-time Fourier transform (STFT), masking techniques, and other methods to suppress electrical noise, heart sounds, and respiratory artifacts. Bowel sound events shorter than 50 ms were excluded; events separated by intervals <100 ms were merged into a single continuous event. The 50-ms threshold was used to reduce very brief impact-like transients that did not permit stable feature extraction, and gaps shorter than 100 ms were treated as interruptions within a single complex acoustic event. These were operational segmentation parameters rather than established physiological boundaries. The resulting clean audio files, together with their waveforms and spectrograms (Figure S2(b)), were independently reviewed by two specialist physicians, who listened to the recordings and identified bowel sound transients and residual artifacts. If they disagreed, a third specialist physician adjudicated the recording. Only validated bowel sounds were included in subsequent analyses. A total of 13 acoustic features were extracted from validated bowel sound events, encompassing time-domain (e.g., event counts, duration, interval, amplitude, zero-crossing rate, short-time energy) and frequency-domain (e.g., dominant, min, and max frequency) characteristics. (Figures S3 and S4). The zero-crossing rate reflects how frequently the waveform crosses the zero-amplitude axis, indicating the oscillation rate and complexity of the signal and showing greater sensitivity to high-frequency or irregular signals. Short-time energy represents the relative acoustic energy or intensity within short time windows, calculated from the mean squared amplitude. Under standardized recording conditions, lower short-time energy indicates weaker overall bowel sound signals. Statistical analysis Continuous variables were tested for normality using the Shapiro-Wilk test. Normally distributed data were compared using the independent samples t-test; otherwise, the Mann-Whitney U test was used. Categorical variables were compared using the Pearson χ2 test, or Fisher's exact test when expected cell counts were <5. Constipation status was assessed before bowel sound recording according to Rome IV criteria, with the medical record used as supporting clinical information. For subgroup analyses, patients were stratified by constipation status (PD with constipation, PD without constipation, and controls) and by H-Y stage (early: H-Y 1-2; moderate-to-severe: H-Y 3-5). Participants were not classified according to the proposed body-first/brain-first framework. For three-group comparisons, normality and homogeneity of variance were assessed using Shapiro-Wilk and Levene's tests, respectively. One-way ANOVA with Tukey's HSD post hoc test was applied when assumptions were met; Welch's ANOVA with Games-Howell post hoc test was used when variances were unequal; and Kruskal-Wallis H test with Dunn's post hoc test (Holm-Bonferroni correction) was used for non-normally distributed data. Two-group comparisons within H-Y subgroups followed the same normality-based strategy (t-test or Mann-Whitney U test). For model development, all acoustic and biochemical features were standardized using Z-scores. Variable selection was performed using LASSO regression with L1 regularization and 10-fold cross-validation to determine the optimal λ. Collinearity was assessed using variance inflation factor (VIF) and correlation matrices; variables with VIF >5 or r > 0.85 were excluded. The remaining LASSO-selected variables were then entered into a multivariable logistic regression model, with age and sex forced into the model a priori to adjust for their potential confounding effects. Model discrimination was assessed using the ROC curve and AUC, and calibration using the Hosmer-Lemeshow goodness-of-fit test (P > 0.05 indicated good calibration). The likelihood ratio test comparing the final model with the null model was used to evaluate overall model significance (P < 0.05). Internal validation was performed using bootstrap resampling with 1000 iterations to calculate optimism-corrected performance metrics. A predictive nomogram was constructed based on the final multivariable model. To evaluate whether bowel sound features alone could provide comparable diagnostic performance, we constructed a reduced model using only the acoustic features selected by LASSO regression, following the identical modeling and validation procedures. Finally, the reduced model was further evaluated in patients with early-stage PD to assess its performance for early diagnosis. Results Demographics A total of 128 PD patients and 105 controls were enrolled. After excluding recordings with premature stethoscope detachment (PD: n = 7; controls: n = 3) or battery depletion (PD: n = 1; controls: n = 2), 120 PD patients and 100 controls with complete 5-minute bowel sound recordings were included in the final analysis. Baseline characteristics, including age, sex, and major comorbidities, were well balanced between the two groups (all P > 0.05). PD patients exhibited significantly altered bowel sound features compared with controls: reduced counts per minute, lower mean short-time energy and amplitude, and shortened max duration (all P < 0.01), accompanied by increased dominant frequency, mean zero-crossing rate, mean and max interval times, and min duration (all P < 0.05). Laboratory findings showed that PD patients had lower triglyceride and glycated hemoglobin levels but higher D-dimer levels (all P < 0.05); no other significant differences were observed. (Table 1) Table 1.Characteristics of PD patients and controls.VariablesControl group (n = 100)PD group (n = 120)P valueDemographics:Age (years)(Median (Q1, Q3))69.00 (62.75–73.00)71.00 (66.75–76.00)0.052Sex (Male/Female)49/5163/570.703Hoehn-Yahr stage:H-Y < 3/48H-Y ≥ 3/72Bowel Sounds:Max interval (s)12.6290 (6.8622–23.0663)20.2695 (11.6795–33.6180)<0.001Min interval (s)0.1720 (0.1630–0.1852)0.1755 (0.1660–0.1930)0.053Mean interval (s)1.4365 (1.0720–2.5360)2.2105 (1.4755–3.8197)<0.001Mean absolute amplitude0.0311 (0.0205–0.0465)0.0233 (0.0178–0.0307)<0.001Mean zero-crossing rate (mean ± SD)0.0275 ± 0.00180.0280 ± 0.00140.028Max frequency (Hz)410.3800 (403.6601–417.2518)410.8581 (405.8425–418.5308)0.255Mean short-time energy0.0045 (0.0023–0.0107)0.0034 (0.0020–0.0054)0.003Dominant frequency (Hz)318.4475 (306.7497–332.6850)324.6318 (315.1335–338.3380)0.018Max duration (s)1.6340 (1.1150–3.3485)1.3615 (0.9228–2.1885)0.008Min duration (s)0.0510 (0.0500–0.0520)0.0510 (0.0500–0.0530)0.039Mean duration (s)0.2825 (0.2095–0.4535)0.2685 (0.1893–0.3702)0.086Bowel sounds per minute40.9150 (23.4125–55.0350)26.1700 (14.9000–40.0100)<0.001Min frequency (Hz)226.7715 (205.7892–233.0550)225.7100 (204.5111–230.7531)0.355Laboratory Tests:Lactate2.36 (1.78–2.77)2.50 (1.98–3.06)0.129Serum calcium2.31 (2.25–2.40)2.31 (2.24–2.38)0.745Serum sodium139.75 (137.90–141.65)140.00 (138.90–141.45)0.236Serum potassium3.90 ± 0.363.86 ± 0.280.367Thyrotropin1.98 (0.97–3.17)1.98 (1.27–2.85)0.894D-dimer146.70 (96.90–304.26)216.40 (129.60–426.73)0.007Homocysteine11.86 (9.28–16.36)11.95 (9.77–16.49)0.587Total Cholesterol4.31 ± 1.404.24 ± 1.130.689High-Density Lipoprotein1.29 (1.05–1.62)1.33 (1.12–1.48)0.685Triglyceride1.17 (0.81–1.84)0.96 (0.72–1.22)0.010Hemoglobin127.00 (116.75–137.48)127.00 (115.00–138.25)0.997Mean Corpuscular Volume92.40 (87.76–95.90)92.45 (89.38–95.48)0.624Glycated hemoglobin6.10 (5.70–8.85)5.90 (5.60–6.40)0.026Low-Density Lipoprotein2.40 (1.65–3.08)2.27 (1.69–2.80)0.311Comorbidities:Hypokalemia2 (2.0%)3 (2.5%)1.000Hyperlipidemia3 (3.0%)4 (3.3%)1.000Hypertension44 (44.0%)44 (36.7%)0.333Hyponatremia1 (1.0%)1 (0.8%)1.000Hypocalcemia2 (2.0%)4 (3.3%)0.691Other Factors:Drinking16 (16.0%)20 (16.7%)1.000Smoking22 (22.0%)35 (29.2%)0.292PD: Parkinson's disease; H-Y: Hoehn-Yahr stages; SD: standard deviation. Q1 represents the 25th percentile, and Q3 represents the 75th percentile. P values are shown as exact values (3 decimals); P < 0.001 is shown as <0.001. Characteristics of PD patients and controls. PD: Parkinson's disease; H-Y: Hoehn-Yahr stages; SD: standard deviation. Q1 represents the 25th percentile, and Q3 represents the 75th percentile. P values are shown as exact values (3 decimals); P < 0.001 is shown as <0.001. Subgroup analysis Constipation subgroup We compared bowel sound acoustic features among three groups: PD patients with constipation, PD patients without constipation, and controls. Significant differences were observed across the three groups in bowel sound counts per minute, max duration, mean zero-crossing rate, mean short-time energy, mean absolute amplitude, and mean and max interval time (Table S1, all P < 0.05). Post hoc analysis revealed that for all of these features, there were no significant differences between the two PD subgroups (with vs. without constipation), whereas both PD subgroups differed significantly from controls (Figure 1, Table S2). These findings suggest that bowel sound abnormalities in PD are not primarily driven by constipation but may reflect broader or subclinical enteric dysfunction beyond overt gastrointestinal symptoms. Violin plots of significant bowel sound features in PD with constipation (PD + C), PD without constipation (PD−C), and controls (Ctrl). PD: Parkinson's disease Abs: absolute. Box plots indicate median and interquartile range. *P < 0.05, **P < 0.01, ***P < 0.001, NS, not significant. (a)–(g) illustrate the distinct variations in acoustic signatures among the three cohorts. H-Y staging subgroup We further stratified PD patients by disease severity based on H-Y staging (early: H-Y < 3; moderate-to-severe: H-Y ≥ 3). Comparison of bowel sound acoustic features between the two subgroups revealed significant differences in max duration (P = 0.01) and mean duration (P = 0.007), with both parameters being longer in early-stage patients (Table 2). No significant differences were observed for the remaining acoustic features (all P > 0.05). The progressive shortening of bowel sound duration with advancing H-Y stage suggests that PD-related intestinal dysmotility may worsen over time. Table 2.Demographic and bowel sound characteristics of PD patients.VariablesH-Y < 3 (n = 48)H-Y ≥ 3 (n = 72)P-valueBowel sounds per minute(median (interquartile range))28.445 (15.523–40.010)25.195 (14.810–38.642)0.500Mean duration (s)0.290 (0.226–0.381)0.229 (0.174–0.349)0.007Min duration (s)0.051 (0.05075–0.053)0.051 (0.05–0.053)0.567Max duration (s)1.706 (1.106–2.705)1.151 (0.794–2.031)0.010Dominant frequency (Hz)(mean ± standard deviation)329.770 ± 20.631324.137 ± 18.5320.122Min frequency (Hz)226.056 (206.243–231.353)225.363 (183.608–230.367)0.461Max frequency (Hz)413.476 (408.160–419.882)410.115 (405.240–417.078)0.095Mean zero-crossing rate0.02821 ± 0.001350.02786 ± 0.001430.190Mean short-term energy0.00318 (0.00205–0.00534)0.00346 (0.00205–0.00552)0.545Mean absolute amplitude0.02335 (0.01797–0.0308)0.02367 (0.01774–0.0305)0.783Mean interval time (s)2.061 (1.473–3.596)2.250 (1.551–3.941)0.534Min interval time (s)0.174 (0.167–0.187)0.176 (0.166–0.196)0.702Max interval time (s)19.927 (11.732–33.618)20.479 (11.679–33.446)0.994PD: Parkinson's disease; H-Y: Hoehn-Yahr stage. P < 0.05 was considered statistically significant. Demographic and bowel sound characteristics of PD patients. PD: Parkinson's disease; H-Y: Hoehn-Yahr stage. P < 0.05 was considered statistically significant. Feature selection We performed two separate LASSO regressions for variable selection. The first, including all acoustic and biochemical features, identified 11 significant predictors (log(λ)≈0.768) (Figure S5). The second, using only bowel sound acoustic features, identified 7 predictors (log(λ)≈0.687) (Figure 2). Collinearity assessment using variance inflation factor (VIF > 5) and correlation matrices (r > 0.85) was applied to the LASSO-selected variables in both models to ensure stability; variables with high collinearity were removed accordingly. Cross-validation plots and coefficient paths for both LASSO regressions are shown in Figure S5 and Figure 2, respectively. After screening and selection, Model 1 ultimately included mean short-time energy, mean interval, min frequency, min duration, max duration, max frequency, D-dimer, HDL (High-Density Lipoprotein), TG (Triglyceride), gender and age. Meanwhile, Model 2 retained mean short-time energy, mean interval, min frequency, min duration, max duration, and max frequency. Feature selection (model 2). (a) Cross-validation AUC curve identifying optimal regularization parameter. (b) LASSO coefficient paths illustrating variable shrinkage. (c) Non-zero LASSO coefficients of the selected acoustic features at the chosen λ. (d) Correlation heatmap among selected features. Model development and predictive factors Using the variables selected by LASSO regression, we constructed two multivariable logistic regression models: Model 1 included acoustic features, biochemical indicators, and sex (forced covariate); Model 2 included acoustic features only. In Model 1, three acoustic features emerged as independent predictors: mean short-time energy (OR = 0.329, 95% CI: 0.169–0.491, P < 0.001) and max duration (OR = 0.355, 95% CI: 0.159–0.520, P < 0.001) were protective factors, whereas max frequency (OR = 1.676, 95% CI: 1.087–3.258, P = 0.019) was a risk factor for PD. Among biochemical indicators, triglyceride (OR = 0.209, 95% CI: 0.067–0.347, P < 0.001) and HDL (OR = 0.644, 95% CI: 0.409–0.928, P = 0.025) showed protective effects, while D-dimer (OR = 1.863, 95% CI: 1.153–4.271, P = 0.021) was identified as a risk factor (Table S3; forest plot in Figure S6(a)). In Model 2, three of the six retained acoustic features were statistically significant: higher values of mean short-time energy (OR = 0.359, 95% CI: 0.237–0.506, P < 0.001) and max duration (OR = 0.384, 95% CI: 0.208–0.557, P < 0.001) were associated with a significantly lower risk of PD, whereas higher max frequency (OR = 1.978, 95% CI: 1.371–3.440, P = 0.005) was associated with an increased risk of PD. (Table S4; forest plot in Figure 3(a)). Logistic regression model performance and prediction(model 2). (a) Forest plot showing adjusted odds ratios with 95% confidence intervals. (b) ROC curve illustrating discriminative ability with area under curve. (c) Nomogram for individualized probability estimation. ROC analysis showed that Model 1 achieved an AUC of 0.8659 (95% CI: 0.8108–0.9093) (Figure S6(b)), and Model 2 achieved an AUC of 0.7999 (95% CI: 0.7365–0.8515) (Figure 3(b)). Model 2 performed comparably to the full model despite using only acoustic features. A nomogram based on Model 2 was constructed for clinical application (Figure 3(c)). Model evaluation and validation We evaluated the predictive performance, stability, and clinical utility of both models using multiple approaches. The Hosmer-Lemeshow test indicated good calibration for both Model 1 (χ2 = 1.89, df = 8, P = 0.984; Figure S7(a)) and Model 2 (χ2 = 9.17, df = 8, P = 0.328; Figure S8(a)). Likelihood ratio tests comparing each full model against its null model confirmed that both models significantly outperformed the null model (Model 1: χ2 = 101.7, df = 11, P < 0.001; Model 2: χ2 = 70.03, df = 6, P < 0.001; Figure S7B and S8B). Bootstrap internal validation with 1000 resamples yielded optimism-corrected AUCs of 0.839 (95% CI: 0.7962–0.8886; optimism: 0.027) for Model 1, and 0.784 (95% CI: 0.7315–0.8418; optimism: 0.016) for Model 2, suggesting mild optimism in both models (Figures S7(c) and S8(c)). Decision curve analysis demonstrated that both models provided positive net benefits across a wide range of threshold probabilities (Model 1: 0.01–0.99; Model 2: 0.02–0.99), with max net benefits of 0.551 and 0.550, respectively, indicating potential net benefit within the development sample; external validation is required before clinical use (Figures S7(d) and S8(d)). In the subgroup analysis comprising 48 patients with early-stage PD (H-Y stage 1-2) and 100 controls, the original model achieved an AUC of 0.7275 (95% CI, 0.6437-0.8063). At the fixed probability threshold of 0.50, sensitivity was 0.6875 (68.75%; 95% CI 0.5467–0.8005) and specificity 0.6600 (66.00%; 95% CI 0.5628–0.7454). (Figure S9) Discussion This study describes of a characteristic bowel sound acoustic phenotype in PD patients: compared with controls, PD patients present a fragmented, high-frequency, and low-energy pattern. No statistically significant difference was detected between PD patients with and without constipation, and two duration measures differed between H-Y stage groups. Bowel sound recordings capture gastrointestinal acoustic activity during a brief observation period, whereas constipation reflects bowel symptoms over a longer period. The two measures therefore need not correspond directly. This may show bowel sound abnormalities in PD appear to reflect broader subclinical enteric dysfunction rather than being merely secondary to constipation, and their progressive shortening with advancing H-Y stage suggests that PD-related intestinal dysmotility worsens over the disease course. The acoustic-only model contained six predictors, three of which were statistically significant. This model performed comparably to a full model incorporating all covariates (AUC: 0.7999 vs. 0.8659), demonstrating that bowel sound alterations are an independent and direct acoustic signature of PD-related enteric nervous system dysfunction.Unlike prior studies that merely reported group differences in isolated acoustic parameters, our multi-feature modeling approach establishes bowel sound alterations as a quantifiable and PD-specific acoustic signature with robust discriminative performance. Current PD biomarker research is heavily focused on the α-synuclein SAA. 9 Although highly specific in cerebrospinal fluid, SAA requires invasive sampling and lacks standardization, limiting its use in community screening. Efforts to apply SAA to minimally invasive gastrointestinal biopsies-including gastroduodenal and skin biopsies-have shown encouraging diagnostic performance, 12 yet these approaches still require endoscopic procedures and tissue processing, restricting their scalability for population-based screening. By contrast, bowel sound recording is non-invasive and may provide complementary functional information about gastrointestinal involvement in PD. In the exploratory H-Y 1-2 analysis, the fitted model achieved an AUC of 0.728, indicating that acoustic features retained discriminatory value in patients with mild motor impairment. The timing of gastrointestinal involvement may help explain the presence of these acoustic alterations in early-stage PD. Within the proposed body-first/brain-first framework, the peripheral autonomic or enteric nervous system may be affected before the nigrostriatal motor system in body-first PD.7,8 Such patients may therefore develop gastrointestinal dysmotility and associated bowel sound abnormalities when motor symptoms are still mild or even during the prodromal stage. Conversely, in the brain-first phenotype, the central dopaminergic system may be affected relatively early, whereas peripheral autonomic and gastrointestinal alterations emerge later.7,8 Early bowel sound abnormalities may consequently be milder or less consistent in these patients. This framework highlights the potential value of bowel sound analysis for characterizing gastrointestinal involvement alongside motor assessment and provides a rationale for investigating acoustic differences across PD phenotypes. Unlike previous bowel sound studies in PD, which relied on limited feature sets and small samples without rigorous confounder adjustment, 13 the present study implemented a standardized acoustic protocol with comprehensive covariate control (including age, sex, disease duration, and medication status) and validated the specificity of the acoustic findings through constipation-stratified, H-Y-stratified and early-stage PD patients analyses. This multi-layered analytical approach strengthens the evidence that bowel sound alterations reflect PD-specific enteric neuropathy rather than non-specific gastrointestinal symptoms or medication effects, addressing a critical gap in the existing literature where the specificity of acoustic findings had not been systematically evaluated. Although bowel sound analysis techniques have become relatively mature at the engineering level, the lack of clinically standardized acquisition and processing workflows has limited their widespread application. 14 To address this gap, the present study implemented a rigorous unified protocol-fixed fasting times, standardized recording environments, uniform probe placement, and a clear medication withdrawal window-to minimize confounding bias. These measures provide a reproducible framework for future cross-device and external validation. We envisage bowel sound analysis as a research approach with potential for use as an adjunct to clinical assessment. Mechanistically, previous studies have reported delayed gastrointestinal motility, enteric α-synuclein pathology, intestinal inflammation and barrier dysfunction, small intestinal bacterial overgrowth, and autonomic involvement in PD.15–20 These abnormalities may affect bowel sounds by altering the strength and coordination of intestinal contractions and the movement of intraluminal gas and fluid. In this context, the lower short-time energy observed in patients with PD may be compatible with reduced acoustic energy generated during intestinal activity, whereas the shorter maximum duration may reflect more fragmented or less sustained contractile events. The increase in maximum frequency could arise from brief, localized contractions and changes in gas-liquid flow, which may generate transient high-frequency acoustic components. Autonomic dysfunction, including impaired vagal regulation, together with dysbiosis-related changes in intraluminal gas, may further contribute to this acoustic pattern. Overall, the fragmented, high-frequency, and low-energy phenotype may represent a functional acoustic correlate of altered gastrointestinal motility and coordination in PD. To determine whether acoustic features contributed information beyond routine blood markers, we compared models with and without biochemical variables. The same three acoustic predictors were statistically significant in both models. Associations of triglyceride, HDL, and D-dimer are hypothesis-generating; although previous studies suggest possible links between gut-derived metabolites, α-synuclein models, lipid biology, and coagulation in PD.21–25 Given the cross-sectional design, these biochemical findings are best regarded as exploratory and provide directions for longitudinal mechanistic studies. Computerized bowel sound analysis has been applied to irritable bowel syndrome, Crohn's disease, and inflammatory bowel disease,26–28 highlighting the potential of acoustic phenotyping to capture functional gastrointestinal alterations. The present findings extend this approach to PD and identify a fragmented, high-frequency, and low-energy acoustic phenotype associated with the disease. Standardized acquisition and signal-processing protocols may facilitate future comparisons of this phenotype across PD, primary gastrointestinal disorders, and other neurodegenerative diseases. Limitations The cross-sectional design limits the generalizability of these findings. The early-stage subgroup analysis was conducted within the model-development cohort, and independent multicenter validation is needed to establish performance across clinical settings. The cross-sectional design describes differences across disease stages; longitudinal studies will clarify when bowel sound alterations emerge and how they evolve. Although fasting duration, recording conditions, and medication-withholding periods were standardized, habitual diet, physical activity, and residual medication effects may have influenced the recordings. Controls were recruited from neurology inpatients, and comparisons with community participants, other neurodegenerative disorders, and primary gastrointestinal conditions will help define the disease specificity and clinical applicability of the acoustic phenotype. Conclusions This study identified a characteristic bowel sound acoustic phenotype in PD, marked by fragmented, higher-frequency, and lower-energy signals. Acoustic features did not differ significantly by constipation status, whereas differences in duration measures across H-Y stage groups suggested a relationship with disease severity. The acoustic-only model distinguished patients with PD from controls and retained discriminatory value in patients with mild motor impairment. These findings support the potential of bowel sound analysis as a non-invasive approach to characterizing gastrointestinal involvement in PD and providing complementary information for multimodal assessment. Supplemental Material sj-docx-1-pkn-10.1177_1877718X261499238 - Supplemental material for An acoustic-only multivariable diagnostic model for Parkinson's disease using bowel sound acoustic signatures Supplemental material, sj-docx-1-pkn-10.1177_1877718X261499238 for An acoustic-only multivariable diagnostic model for Parkinson's disease using bowel sound acoustic signatures by Sicheng Chen, Man Zhang, Surui Fang, Qi Yang, Chuyi Chen, Zhixuan Lei, Yuxia Zhu, Fen Deng, Lijuan Mo, Xi Liu, Changhong Tan and Lifen Chen in Journal of Parkinson's Disease